Dissertations / Theses on the topic 'Fault diagnosis'

The modern VLSI circuit designs manufactured with advanced technology nodes of 65nm or below exhibit an increasing sensitivity to the variations of manufacturing process. New design-specific and feature-sensitive failure mechanisms are on the rise. Systematic yield issues can be severe due to the complex variability involved in process and layout features. Without improved yield analysis methods, time-to-market is delayed, mature yield is suboptimal, and product quality may suffer, thereby undermining the profitability of the semiconductor company. Diagnosis-driven yield improvement is a methodology that leverages production test results, diagnosis results, and statistical analysis to identify the root cause of yield loss and fix the yield limiters to improve the yield. To fully leverage fault diagnosis, the diagnosis-driven yield analysis requires that the diagnosis tool should provide high-quality diagnosis results in terms of accuracy and resolution. In other words, the diagnosis tool should report the real defect location without too much ambiguity. The second requirement for fast diagnosis-driven yield improvement is that the diagnosis tool should have the capability of processing a volume of failing dies within a reasonable time so that the statistical analysis can have enough information to identify the systematic yield issues. In this dissertation, we first propose a method to accurately diagnose the defects inside the library cells when multi-cycle test patterns are used. The methods to diagnose the interconnect defect have been well studied for many years and are successfully practiced in industry. However, for process technology at 90nm or 65nm or below, there is a significant number of manufacturing defects and systematic yield limiters lie inside library cells. The existing cell internal diagnosis methods work well when only combinational test patterns are used, while the accuracy drops dramatically with multi-cycle test patterns. A method to accurately identify the defective cell as well as the failing conditions is presented. The accuracy can be improved up to 94% compared with about 75% accuracy for previous proposed cell internal diagnosis methods. The next part of this dissertation addresses the throughput problem for diagnosing a volume of failing chips with high transistor counts. We first propose a static design partitioning method to reduce the memory footprint of volume diagnosis. A design is statically partitioned into several smaller sub-circuits, and then the diagnosis is performed only on the smaller sub-circuits. By doing this, the memory usage for processing the smaller sub-circuit can be reduced and the throughput can be improved. We next present a dynamic design partitioning method to improve the throughput and minimize the impact on diagnosis accuracy and resolution. The proposed dynamic design partitioning method is failure dependent, in other words, each failure file has its own design partition. Extensive experiments have been designed to demonstrate the efficiency of the proposed dynamic partitioning method.

Fault tree analysis is a method that describes all possible causes of a specified system state in terms of the state of the components within the system. Fault trees are commonly developed to analyse the adequacy of systems, from a reliability or safety point of view during the stages of design. The aim of the research presented in this thesis was to develop a method for diagnosing faults in systems using a model-based fault tree analysis approach, taking into consideration the potential for use on aircraft systems. Initial investigations have been conducted by developing four schemes that use coherent and non-coherent fault trees, the concepts of which are illustrated by applying the techniques to a simple system. These were used to consider aspects of system performance for each scheme at specified points in time. The results obtained were analysed and a critical appraisal of the findings carried out to determine the individual effectiveness of each scheme. A number of issues were highlighted from the first part of research, including the need to consider dynamics of the system to improve the method. The most effective scheme from the initial investigations was extended to take into account system dynamics through the development of a pattern recognition technique. Transient effects, including time history of flows and rate of change of fluid level were considered. The established method was then applied to a theoretical version of the BAE Systems fuel rig to investigate how the method could be utilised on a larger system. The fault detection was adapted to work with an increased number of fuel tanks and other components adding to the system complexity. The implications of expanding the method to larger systems such as a full aircraft fuel system were identified for the Nimrod MRA4.

Le nombre de circuits intégrés (CIs) utilisés dans les applications liées à des missions critiques et à la sûreté augmente sans cesse. Ces applications imposent aux CIs de présenter des propriétés de sûreté fonctionnelle. Cette thèse introduit un auto-test intégré (BIST) pour les CIs analogiques et à signaux mixtes, appelé autotest à symétrie (SymBIST) pour répondre à l’objectif de sûreté fonctionnelle. SymBIST repose sur le principe du BIST et sur l'existence de signaux invariants en fonctionnement nominal et variant en cas de fonctionnement erroné. Les invariants sont mesurés à l'aide de dispositifs intégrés spécifiques. SymBIST répond à trois objectifs de sûreté fonctionnelle : le test les défauts du CI, le test en ligne, et le diagnostic les défauts. SymBIST est démontré sur un convertisseur analogique-numérique à approximations successives (CAN SAR). Les résultats montent que la couverture de test et la précision de diagnostic sont plus élevées que l’état de l’art
The number of integrated circuits (ICs) used in safety- and mission-critical applications is ever increasing. These applications demand that ICs carry functional safety properties. In this thesis, we develop a Built-In Self Test (BIST) approach for Analog and Mixed-Signal (A/M-S) ICs, called Symmetry-Based Built-In Self Test (SymBIST), which achieves several objectives towards the functional safety goal. SymBIST is a generic BIST paradigm based on identifying inherent invariances that should hold true only in error-free operation, while their violation points to abnormal operation. The invariances are being checked using dedicated on-die checkers. SymBIST meets three functional safety objectives: post-manufacturing defect-oriented test, on-line testing, and fault diagnosis. SymBIST is demonstrated on a successive approximation analog-to-digital converter (SAR ADC). The results show that the test coverage and diagnostic accuracy are promising compared to the state of the art

In this thesis, topics of importance to the fault diagnosis of rotating machinery in the power generation industry have been addressed, including a review of the relevant literature and an overview of the associated rotordynamics modelling and analysis techniques. For faults involving rotor-stator interaction it has been shown that the inclusion of torsion in mathematical models used for rotor-stator contract analyses can have a significant influence on the dynamic behaviour of the system. A 3 degrees-of-freedom model based on the Jeffcott rotor was developed and, for physically realistic systems, it was shown that very different results were obtained when including torsion, compared to when torsion was neglected, as has generally been the case in the past. An identification method for estimating both the excitation and flexible support parameters of a rotor-bearings-foundations system has been presented. Excitation due to both mass unbalance and a bent rotor were included in the analysis, which has been verified both in simulation and experimentally. The method has great practical potential, since it allows balancing to be performed using data obtained from just a single run-up or run-down, which has obvious benefits for field balancing. Using this single-shot balancing technique in experiment, vibration levels were successfully reduced by as much as 92% of their original levels. A bent rotor has been accurately identified in both simulation and experiment. It was also shown that including bend identification in those cases where only unbalance forcing was present in no way detracted from the accuracy of the estimated unbalance or foundation parameters. The identification of the flexible foundation parameters was generally successful, with measured and estimated parameters matching very closely in most cases. The identification method was tested for a wide range of conditions and proved suitably robust to changes in the system configuration, noisy data and modelling error.

The aim of this thesis is to contribute to improved diagnosis of automotive vehicles. The work is driven by case studies, where problems and challenges are identified. To solve these problems, theoretically sound and general methods are developed. The methods are then applied to the real world systems. To fulfill performance requirements automotive vehicles are becoming increasingly complex products. This makes them more difficult to diagnose. At the same time, the requirements on the diagnosis itself are steadily increasing. Environmental legislation requires that smaller deviations from specified operation must be detected earlier. More accurate diagnostic methods can be used to reduce maintenance costs and increase uptime. Improved diagnosis can also reduce safety risks related to vehicle operation. Fault diagnosis is the task of identifying possible faults given current observations from the systems. To do this, the internal relations between observations and faults must be identified. In complex systems, such as automotive vehicles, finding these relations is a most challenging problem due to several sources of uncertainty. Observations from the system are often hidden in considerable levels of noise. The systems are complicated to model both since they are complex and since they are operated in continuously changing surroundings. Furthermore, since faults typically are rare, and sometimes never described, it is often difficult to get hold of enough data to learn the relations from. Due to the several sources of uncertainty in fault diagnosis of automotive systems, a probabilistic approach is used, both to find the internal relations, and to identify the faults possibly present in the system given the current observations. To do this successfully, all available information is integrated in the computations. Both on-board and off-board diagnosis are considered. The two tasks may seem different in nature: on-board diagnosis is performed without human integration, while the off-board diagnosis is mainly based on the interactivity with a mechanic. On the other hand, both tasks regard the same vehicle, and information from the on-board diagnosis system may be useful also for off-board diagnosis. The probabilistic methods are general, and it is natural to consider both tasks. The thesis contributes in three main areas. First, in Paper 1 and 2, methods are developed for combining training data and expert knowledge of different kinds to compute probabilities for faults. These methods are primarily developed with on-board diagnosis in mind, but are also applicable to off-board diagnosis. The methods are general, and can be used not only in diagnosis of technical system, but also in many other applications, including medical diagnosis and econometrics, where both data and expert knowledge are present. The second area concerns inference in off-board diagnosis and troubleshooting, and the contribution consists in the methods developed in Paper 3 and 4. The methods handle probability computations in systems subject to external interventions, and in particular systems that include both instantaneous and non-instantaneous dependencies. They are based on the theory of Bayesian networks, and include event-driven non-stationary dynamic Bayesian networks (nsDBN) and an efficient inference algorithm for troubleshooting based on static Bayesian networks. The framework of nsDBN event-driven nsDBN is applicable to all kinds of problems concerning inference under external interventions. The third contribution area is Bayesian learning from data in the diagnosis application. The contribution is the comparison and evaluation of five Bayesian methods for learning in fault diagnosis in Paper 5. The special challenges in diagnosis related to learning from data are considered. It is shown how the five methods should be tailored to be applicable to fault diagnosis problems. To summarize, the five papers in the thesis have shown how several challenges in automotive diagnosis can be handled by using probabilistic methods. Handling such challenges with probabilistic methods has a great potential. The probabilistic methods provide a framework for utilizing all information available, also if it is in different forms and. The probabilities computed can be combined with decision theoretic methods to determine the appropriate action after the discovery of reduced system functionality due to faults.

With the increased use of power electronics in aerospace, automotive, industrial, and energy generation sectors, the demand for highly reliable and power dense solutions has increased. Taking into account the demands for high reliability and high power density, matrix converters become attractive. With their lack of large bulky DC- Link capacitors, high power densities are possible with capability to operate with high ambient temperatures [7]. Demand for high reliability under tight weight and volume constrains, often it is not possible to have an entirely redundant system. Under these conditions it is desirable that the system continue to operate even under faulty conditions, albeit with diminished performance in some regard. Research has been carried out on the continued operation of a matrix converter during an open- circuit switch failure[8][9]. These methods however assume that a fault detection and diagnosis system was already in place. The behavior of matrix converters under fault conditions are more complex than traditional inverter drive systems, as there is no decoupling through the DC-Link and the matrix converter's clamp circuit also complicates matters. This thesis describes the operation of a matrix converter and the clamp circuit during a open-circuit fault condition and presents a number of methods for fault detection and diagnosis in matrix converters.

The development of decision support systems for equipment diagnosis has been found to be an iterative process whereby functionality and knowledge are continually added to a prototype until satisfactory performance is achieved. In order to reduce both the dependency on compiled knowledge sources and the number of prototype stages necessary to develop diagnostic decision support systems, this thesis examines, adapts and applies a set theoretical approach to mechanism diagnosis first developed in the field of Artificial Intelligence. The approach does not require the development of computational models to simulate equipment behaviour.
The set theoretical approach was applied to the development of a diagnostic decision support system for a semi-automated Atlas Copco Wagner ST-8B Load-Haul-Dump vehicle. Hypothesis sets were generated for the vehicle's hydraulic circuit and Deutz FL-413-FW diesel engine. A high level of diagnostic resolution was achieved for the hydraulic circuit, but limited resolution was achieved for the diesel engine. This was postulated to be due to the ratio of observable system outputs to input sub-systems, and the number of least repairable units making up each system.
Manual knowledge acquisition was undertaken in an underground mine to refine the diagnostic knowledge developed from the hypothesis sets and to add knowledge to discriminate between competing failure hypotheses. Heuristic failure likelihoods were used to rank hypotheses in order of frequency of occurrence. The knowledge base was implemented as a hypertext decision support system using HyperText Mark-up Language (HTML). The resulting decision support system is platform independent, upgradeable and able to be maintained by site personnel. The system is currently installed at surface level and at 1800 level at INCO Limited's Stobie Mine in Sudbury, Ontario.
The thesis makes a number of original contributions, the first two of which are of generic significance. It is the first work to apply set theoretical concepts to structural models of mobile mining equipment in order to diagnose faults. A number of modifications are advanced to the conventional trace-back analysis technique for generating contributor and normality sets, and heuristic guidelines are provided for estimating the costs and benefits of developing, implementing and maintaining diagnostic decision support systems. It is also the first work to formalise a decision support system in HTML and to suggest the application of company-wide internets ("intranets") to disseminate maintenance knowledge within mines.

Fault detection and isolation (FDI) has become one of the most important aspects of automobile design. In this thesis, a new fault detection and isolation approach is developed for automotive engines. The method uses an independent radial basis function neural network model to model engine dynamics, and the modelling errors are used to from the basis for residual generation. Furthermore, another radial basis function neural network is used as a fault classifier to isolate an occurred fault from other possible faults in the system by classifying fault characteristics embedded in the modelling errors. The performance of the developed scheme is assessed using an engine benchmark, the Mean Value Engine Model CMVEM), with Matlab/Simulink. Five faults have been simulated on the MVEM: three sensor faults, one component fault and one actuator fault. The three sensor faults considered are 10~20% changes superimposed on the measured outputs of manifold pressure, manifold temperature and crankshaft speed sensors; the component fault considered is air leakage in the intake manifold; and the actuator fault considered is the malfunction of the fuel injector. The simulation results show that all the simulated faults can be clearly detected and isolated in dynamic conditions throughout the engine operating range. Furthermore, in order to reflect the real state for an automotive engine, the FDI method is evaluated for the MVEM system under closed-loop control with air fuel ratio control. An independent radial basis function (RBF) neural network model is used to model engine dynamics using random amplitude signals (RAS) throttle angle as an input.

A "causal" expert system based on hypothetical reasoning and its application to a Mark 45 turret gun's lower hoist are described. HOIST is a system that performs fault diagnosis without the use of a domain expert or "shallow rules". Rather its "knowledge" is coded directly from a structural specification of the Mark 45 lower hoist. The technology reported here for assisting the lesser acquainted diagnostician differs considerably from the normal rule-based expert system techniques: it reasons about machine failures from a functional model of the device. In a mechanism like the lower hoist, a functional model must reason about forces, fluid pressures and mechanical linkages, that is, qualitative physics. HOIST technology can be directly applied to any exactly specified device for modeling and diagnosis of single or multiple faults. Hypothetical reasoning, the process embodied in HOIST, has general utility in qualitative physics and reason maintenance.

Master of Science

Operators can become confused while diagnosing faults in process plant while in operation. This may prevent remedial actions being taken before hazardous consequences can occur. The work in this thesis proposes a method to aid plant operators in systematically finding the causes of any fault in the process plant. A computer aided fault diagnosis package has been developed for use on the widely available IBM PC compatible microcomputer. The program displays a coloured diagram of a fault tree on the VDU of the microcomputer, so that the operator can see the link between the fault and its causes. The consequences of the fault and the causes of the fault are also shown to provide a warning of what may happen if the fault is not remedied. The cause and effect data needed by the package are obtained from a hazard and operability (HAZOP) study on the process plant. The result of the HAZOP study is recorded as cause and symptom equations which are translated into a data structure and stored in the computer as a file for the package to access. Probability values are assigned to the events that constitute the basic causes of any deviation. From these probability values, the a priori probabilities of occurrence of other events are evaluated. A top-down recursive algorithm, called TDRA, for evaluating the probability of every event in a fault tree has been developed. From the a priori probabilities, the conditional probabilities of the causes of the fault are then evaluated using Bayes' conditional probability theorem. The posteriori probability values could then be used by the operators to check in an orderly manner the cause of the fault. The package has been tested using the results of a HAZOP study on a pilot distillation plant. The results from the test show how easy it is to trace the chain of events that leads to the primary cause of a fault. This method could be applied in a real process environment.

Bridge fault extractors are tools that analyze chip layouts and produce a realistic list of bridging faults within that chip. FedEx, previously developed at Texas A&M University, extracts all two-node intralayer bridges of any given chip layout and optionally extracts all two-node interlayer bridges. The goal of this thesis was to further develop this tool. The primary goal was to speed it up so that it can handle large industrial designs in a reasonable amount of time. A second goal was to develop a graphical user interface (GUI) for this tool which aids in more effectively visualizing the bridge faults across the chip. The final aim of this thesis was to perform FedEx output analysis to understand the nature of the defects, such as variation of critical area (the area where the presence of a defect can cause a fault) as a function of layer as well as defect size.

Nell’ultimo decennio, la crescente richiesta di produzione di energia elettrica da fonti rinnovabili, ha generato una cospicua attenzione nei riguardi delle turbine eoliche. Si tratta di sistemi particolarmente complessi, che richiedono affidabilit`a, sicurezza, manutenzione e, soprattutto, efficienza nella produzione di potenza elettrica. Pertanto, sono sorte nuove sfide nel campo della ricerca e sviluppo, in particolare nel contesto della modellazione e del controllo. Sistemi di controllo sostenibile e all’avanguardia possono ottimizzare la conversione di energia e garantire determinate prestazioni, anche in presenza di condizioni di lavoro anomale, causate da malfunzionamenti e guasti inaspettati. Questa tesi tratta la tematica della diagnosi dei guasti e del controllo tollerante al guasto applicato alle turbine eoliche. Si propongono originali soluzioni relative al problema della pronta rivelazione del guasto e del suo trattamento. Il sistema di controllo che si `e sviluppato `e principalmente basato su un modulo di diagnosi del guasto, che ha il compito di fornire in tempo reale l’informazione sull’eventuale guasto presente, in modo da compensare l’azione di controllo. Il progetto degli stimatori di guasto riguarda strategie basate sui dati, poich´e offrono un efficace strumento per la gestione di sistemi le cui dinamiche sono scarsamente conosciute in termini analitici e presentano rumore e disturbi. Il primo di questi approcci basati sui dati `e ottenuto tramite modelli fuzzy Takagi-Sugeno (TS), derivanti dall’algoritmo di clustering c-means, seguito da una procedura di identificazione dei parametetri che risolve il problema della reiezione dei disturbi. Il secondo metodo proposto si serve di reti neurali artificiali per descrivere le relazioni fortemente non lineari che sussistono fra misure e guasti. L’architettura scelta fa parte della topologia Non lineare Autoregressiva con ingresso esogeno (NARX), dato che pu`o rappresentare l’evoluzione dinamica di un sistema nel tempo. L’addestramento della rete neurale sfrutta l’algoritmo di Levenberg-Marquardt con backpropagation, e processa un insieme di dati-obiettivo direttamente acquisiti. Gli schemi di diagnosi del guasto e controllo tollerante al guasto sono stati testati per mezzo di due modelli benchmark ad alta fedelt`a, i quali simulano rispettivamente il comportamento di una singola turbina e di un parco eolico, sia in condizioni normali, sia di guasto. Le prestazioni ottenute sono state confrontate con quelle di altre strategie di controllo, proposte in letteratura. Inoltre, un’analisi Monte Carlo ha validato la robustezza dei sistemi sviluppati, relativa a tipiche variazioni nei parametri, disturbi e incertezze. 1 2 Infine, si `e effettuato un test Hardware In the Loop (HIL), al fine di valutare le prestazioni in un contesto piu` realistico e real-time. L’efficacia mostrata dai risultati ottenuti suggerisce future ricerche sull’effettiva applicabilit`a industriale dei sistemi proposti.
In recent years, the increasing demand for energy generation from renewable sources has led to a growing attention on wind turbines. Indeed, they represent very complex systems which require reliability, availability, maintainability, safety and, above all, efficiency on the generation of electrical power. Thus, new research challenges arise, in particular in the context of modeling and control. Advanced sustainable control systems can provide the optimization of energy conversion and guarantee the desired performances even in presence of possible anomalous working condition, caused by unexpected faults and malfunctions. This thesis deals with the fault diagnosis and the fault tolerant control of wind turbines, and it proposes novel solutions to the problem of earlier fault detection and accommodation. The developed fault tolerant controller is mainly based on a fault diagnosis module, that provides the on-line information on the faulty or fault-free status of the system, so that the controller action can be compensated. The design of the fault estimators involves data-driven approaches, as they offer an effective tool for coping with a poor analytical knowledge of the system dynamics, together with noise and disturbances. The first data-driven proposed solution relies on fuzzy Takagi-Sugeno (TS) models, that are derived from a clustering c-means algorithm, followed by an identification procedure solving the noise-rejection problem. Then, a second solution makes use of neural networks to describe the strongly nonlinear relationships between measurement and faults. The chosen network architecture belongs to the Nonlinear AutoRegressive with eXogenous input (NARX) topology, as it can represent a dynamic evolution of the system along time. The training of the neural network fault estimators exploits the backpropagation Levenberg-Marquardt algorithm, that processes a set of acquired target data. The developed fault diagnosis and fault tolerant control schemes are tested by means of two high-fidelity benchmark models, that simulate the normal and the faulty behavior of a single wind turbine and a wind farm, respectively. The achieved performances are compared with those of other control strategies, coming from the related literature. Moreover, a Monte Carlo analysis validates the robustness of the proposed systems against the typical parameter uncertainties and disturbances. Finally, the Hardware In the Loop (HIL) test is carried out, in order to assess the performance in a more realistic real-time framework. The effectiveness shown by the achieved results suggests further investigations on the industrial application of the proposed systems.

In this dissertation, we explore the problem of fault detection and fault diagnosis for systems modeled as condition systems. A condition system is a Petri net based framework of components which interact with each other and the external environment through the use of condition signals. First, a system FAULT is defined as an observed behavior which does not correspond to any expected behavior, where the expected behavior is defined through condition system models. A DETECTION is the determination that the system is not behaving as expected according to the model of the system. A DIAGNOSIS of this fault localizes the subsystem that is the source of the discrepancy between output and expected observations. We characterize faults as a behavior relaxation of model components. We then show that detection and diagnosis can be determined in a finite number of calculations. The exact solution can be computationally involved, so we also present methods to perform a rapid detection and diagnosis. We have also included a chapter on a conversion from the condition system framework into a linear-time temporal logic(LTL) framework.

Models are of great interest in many fields of engineering as they enable prediction of a systems behaviour, given an initial mode of the system. However, in the field of model-based diagnosis the models are used in a reverse manner, as they are combined with the observations of the systems behaviour in order to estimate the system mode. This thesis describes computation of diagnostic systems based on models implemented in Dymola. Dymola is a program that uses the language Modelica. The Dymola models are translated to Matlab, where an application called Fault Diagnosis Toolbox, FDT is applied. The FDT has functionality for pinpointing minimal overdetermined sets of equations, MSOs, which is developed further in this thesis. It is shown that the implemented algorithm has exponential time complexity with regards to what level the system is overdetermined,also known as the degree of redundancy. The MSOs are used to generate residuals, which are functions that are equal to zero given that the system is fault-free. Residual generation in Dymola is added to the original methods of the FDT andthe results of the Dymola methods are compared to the original FDT methods, when given identical data. Based on these tests it is concluded that adding the Dymola methods to the FDT results in higher accuracy, as well as a new way tocompute optimal observer gain. The FDT methods are applied to 2 models, one model is based on a system ofJAS 39 Gripen; SECS, which stands for Secondary Enviromental Control System. Also, applications are made on a simpler model; a Two Tank System. It is validated that the computational properties of the developed methods in Dymolaand Matlab differs and that it therefore exists benefits of adding the Dymola implementations to the current FDT methods. Furthermore, the investigation of the potential isolability based on the current setup of sensors in SECS shows that full isolability is achievable by adding 2 mass flow sensors, and that the isolability is not limited by causality constraints. One of the found MSOs is solvable in Dymola when given data from a fault-free simulation. However, if the simulation is not fault-free, the same MSO results in a singular equation system. By utilizing MSOs that had no reaction to any modelled faults, certain non-monitored faults is isolated from the monitored ones and therefore the risk of false alarms is reduced. Some residuals are generated as observers, and a new method for constructing observers is found during the thesis by using Lannerheds theorem in combination with Pontryagin’s Minimum Priniple. This method enables evaluation of observer based residuals in Dymola without any selection of a specific operating point, as well as evaluation of observers based on high-index Differential Algebraic Equations, DAEs. The method also results in completely different behaviourof the estimation error compared to the method that is already implemented inthe FDT. For example, one of the new observer-implementations achieves both an estimation error that converges faster towards zero when no faults are implementedin the monitored system, and a sharper reaction to implemented faults.

With the development of the embedded application and driving assistance systems, it becomes relevant to develop parallel mechanisms in order to check and to diagnose these new systems. In this thesis we focus our research on one of this type of parallel mechanisms and analytical redundancy for fault diagnosis of an automotive suspension system. We have considered a quarter model car passive suspension model and used a parameter estimation, ARX model, method to detect the fault happening in the damper and spring of system. Moreover, afterward we have deployed a neural network classifier to isolate the faults and identifies where the fault is happening. Then in this regard, the safety measurements and redundancies can take into the effect to prevent failure in the system. It is shown that The ARX estimator could quickly detect the fault online using the vertical acceleration and displacement sensor data which are common sensors in nowadays vehicles. Hence, the clear divergence is the ARX response make it easy to deploy a threshold to give alarm to the intelligent system of vehicle and the neural classifier can quickly show the place of fault occurrence.

El desalineamiento temporal es la incorrespondencia de dos señales debido a una distorsión en el eje temporal. La Detección y Diagnóstico de Fallas (Fault Detection and Diagnosis-FDD) permite la detección, el diagnóstico y la corrección de fallos en un proceso. La metodología usada en FDD está dividida en dos categorías: técnicas basadas en modelos y no basadas en modelos. Esta tesis doctoral trata sobre el estudio del efecto del desalineamiento temporal en FDD. Nuestra atención se enfoca en el análisis y el diseño de sistemas FDD en caso de problemas de comunicación de datos, como retardos y pérdidas. Se proponen dos técnicas para reducir estos problemas: una basada en programación dinámica y la otra en optimización. Los métodos propuestos han sido validados sobre diferentes sistemas dinámicos: control de posición de un motor de corriente continua, una planta de laboratorio y un problema de sistemas eléctricos conocido como hueco de tensión.
Time misalignment is the unmatching of two signals due to a distortion in the time axis. Fault Detection and Diagnosis (FDD) deals with the timely detection, diagnosis and correction of abnormal conditions of faults in a process. The methodology used in FDD is clearly dependent on the process and the sort of available information and it is divided in two categories: model-based and non-model based techniques. This doctoral dissertation deals with the study of time misalignments effects when performing FDD. Our attention is focused on the analysis and design of FDD systems in case of data communication problems, such as delays and dropouts. Techniques based on dynamic programming and optimization are proposed to deal with these problems. Numerical validation of the proposed methods is performed on different dynamic systems: a control position for a DC motor, a laboratory plant and an electrical system problem known as voltage sag.

In a system like a Scania heavy duty truck, faultcodes (DTCs) are generated and stored locally in the ECUs when components, e.g. sensors or actuators, malfunction. Tests are run periodically to detect failure in the system. The test results are processed by the diagnostic system that tries to isolate the faulty components and set local faultcodes.

Currently, in a Scania truck, local diagnoses are only based on local diagnostic information, which the DTCs are based upon. The diagnosis statement can, however, be more complete if diagnoses from other ECUs are considered. Thus a system that extends the local diagnoses by exchanging diagnostic information between the ECUs is desired. The diagnostic information to share and how it should be done is elaborated in this thesis. Further, a model of distributed diagnosis is given and a few distributed diagnostic algorithms for transmitting and receiving diagnostic information are presented.

A basic idea that has influenced the project is to make the diagnostic system scalable with respect to hardware and thereby making it easy to add and remove ECUs. When implementing a distributed diagnostic system in networked real-time embedded systems, technical problems arise such as memory handling, process synchronization and transmission of diagnostic data and these will be discussed in detail. Implementation of a distributed diagnostic system is further complicated due to the fact that the isolation process is a non deterministic job and requires a non deterministic amount of memory.

This dissertation presents a methodology for an efficient, non-destructive, and online incipient fault diagnosis system (IFDS) to detect underground cable incipient faults before they become catastrophic. The system provides vital information to help the operator with the decision-making process regarding the condition assessment of the underground cable. It incorporates advanced digital signal processing and pattern recognition methods to classify recorded data into designated classes. Additionally, the IFDS utilizes novel detection methodologies to detect when the cable is near failure. The classification functionality is achieved through employing an ensemble of rule-based and supervised classifiers. The Support Vector Machines, designed and used as a supervised classifier, was found to perform superior. In addition to the normalized energy features computed from wavelet packet analysis, two new features, namely Horizontal Severity Index, and Vertical Severity Index are defined and used in the classification problem. The detection functionality of the IFDS is achieved through incorporating a temporal severity measure and a detection method. The novel severity measure is based on the temporal analysis of arrival times of incipient abnormalities, which gives rise to a numeric index called the Global Severity Index (GSI). This index portrays the progressive degradation path of underground cable as catastrophic failure time approaches. The detection approach utilizes the numerical modeling capabilities of SOM as well as statistical change detection techniques. The natural logarithm of the chronologically ordered minimum modeling errors, computed from exposing feature vectors to a trained SOM, is used as the detection index. Three modified change detection algorithms, namely Cumulative Sum, Exponentially Weighted Moving Averages, and Generalized Likelihood Ratio, are introduced and applied to this application. These algorithms determine the change point or near failure time of cable from the instantaneous values of the detection index. Performance studies using field recorded data were conducted at three warning levels to assess the capability of the IFDS in predicting the faults that actually occurred in the monitored underground cable. The IFDS presents a high classification rate and satisfactory detection capability at each warning level. Specifically, it demonstrates that at least one detection technique successfully provides an early warning that a fault is imminent.

To be able to do the correct action when a fault is detected, the fault isolation part must be precise and run in real time during operation of the process. In many cases can it be difficult to decide exactly where the fault is localized. In those cases, the isolation algorithm must rank the faults according to their probability to be the cause to the behavior. The masters thesis project aims at probabilistic methods and algorithms for fault isolation in embedded systems. Different kind of Bayesian Networks have been compared in this report and the comparison has been done on a literature defined “benchmark system”. Those Bayesian network models which have been implemented for fault isolation are: 1. Manually (on the basis of physical representations) 2. Two-layer structure continuous signals discreet signals 3. Via temporal causal graph (dynamical network) The algorithms should be compared in the following areas: computational complexity, isolation performance and degree of difficulty to construct the network on the basis of data. The evaluated algorithms showed good results. Even though the system data which have been used in the Bayesian Networks are not very accurate in the first place, it manage to give a fairly precise isolation of the faults. The continuous Bayesian Network manage to show a good isolation performance for different type of faults and the Dynamic Bayesian Network found most of the faults even for a rather complex network.
Detta examensarbete handlar om sannolikhetsbaserade metoder för felisolering. När ett fel uppstår ombord på en Scania lastbil kan man upptäcka det. I bästa fall kan en viss komponent pekas ut som orsak, men ofta kommer man att ha ett antal komponenter som kan vara orsaken. I många fall är det dock svårt att hitta var exakta felet finns. För att hantera dessa situationer vill man använda metoder för att beräkna sannolikheten att olika komponenter är trasiga. För att beräkna sannolikheten kan man använda en probabilistisk model, dvs. Bayesianska nätverk. I detta arbete har olika metoder för att skapa Bayesianska nätverk jämförts. Jämförelsen görs på ett litteratur väl definierat benchmark problem: diagnosar en två-tank system. De typer av Bayesiansk nätverks modeller som har implementerats för felisolering är: 1. Manuellt (ut ifrån fysikalisk modell) 2. Två-lagers struktur kontinuerliga signaler diskreta signaler 3. Via Bindningsgrafer (dynamiskt nätverk) Problemen som undersöktes var bland annat svårighet att bygga nätverket utifrån data, beräkningskomplexitet samt isolerings prestanda. En jämförelse mellan de Bayesianska metoderna för felisolering och samt dem befintliga standardmetoder har även gjorts. De undersökta algoritmerna visade goda resultat. Trots bristen på data, visade algoritmerna lovande resultat. Det Två-lagers Bayesianska nätverket visade en bra isoleringsprestanda på olika komponent fel och det Dynamiska Bayesianska nätverket upptäckte de flesta fel trots att det var ett ganska complext nätverk.

Testing digital devices constitutes a major portion of the cost and effort involved in their design, production, and use. Built-In Self-Test (BIST) has been warmly embraced by the Integrated Circuits (IC) industry as a solution for the continuously aggravating testing problem. BIST provides a simple go/no-go test screening answer. However, when test fall-out is high, it becomes necessary to diagnose faults to improve the yield.
Signature Analysis (SA) is typically used in a BIST environment to compact the outputs of a module into a final signature. Several SA-based diagnostic schemes have been developed in the past. An overwhelming majority of these techniques assume the presence of very few error bits in the Test Response Sequence (TRS). However, this assumption is generally unrealistic since a faulty device in a practical BIST environment can generate an enormous number of erroneous bits in the TRS.
In this thesis, a comprehensive survey of the current SA-based BIST diagnostic schemes is presented first. Then, novel BIST fault diagnosis techniques for scan-based VLSI modules are presented, based on multiple signature analysis. (Abstract shortened by UMI.)

Substantial economic and even safety related gains can be achieved if effective gas turbine performance analysis is attained. During the development phase, analysis can help understand the effect on the various components and on the overall engine performance of the modifications applied. During usage, analysis plays a major role in the assessment of the health status of the engine. Both condition monitoring of operating engines and pass off tests heavily rely on the analysis. In spite of its relevance, accurate performance analysis is still difficult to achieve. A major cause of this is measurement uncertainty: gas turbine measurements are affected by noise and biases. The simultaneous presence of engine and sensor faults makes it hard to establish the actual condition of the engine components. To date, most estimation techniques used to cope with measurement uncertainty are based on Kalman filtering. This classic estimation technique, though, is definitely not effective enough. Typical Kalman filter results can be strongly misleading so that even the application of performance analysis may become questionable. The main engine manufactures, in conjunction with research teams, have devised modified Kalman filter based techniques to overcome the most common drawbacks. Nonetheless, the proposed methods are not able to produce accurate and reliable performance analysis. In the present work a different approach has been pursued and a novel method developed, which is able to quantify the performance parameter variations expressing the component faults in presence of noise and a significant number of sensor faults. The statistical basis of the method is sound: the only accepted statistical assumption regards the well known measurement noise standard deviations. The technique is based on an optimisation procedure carried out by means of a problem specific, real coded Genetic Algorithm. The optimisation based method enables to concentrate the steady state analysis on the faulty engine component(s). A clear indication is given as to which component(s) is(are) responsible for the loss of performance. The optimisation automatically carries out multiple sensor failure detection, isolation and accommodation. The noise and biases affecting the parameters setting the operating point of the engine are coped with as well. The technique has been explicitly developed for development engine test bed analysis, where the instrumentation set is usually rather comprehensive. In other diagnostic cases (pass off tests, ground based analysis of on wing engines), though, just few sensors may be present. For these situations, the standard method has been modified to perform multiple operating point analysis, whereby the amount of information is maximised by simultaneous analysis of more than a single test point. Even in this case, the results are very accurate. In the quest for techniques able to cope with measurement uncertainty, Neural Networks have been considered as well. A novel Auto-Associative Neural Network has been devised, which is able to carry out accurate sensor failure detection and isolation. Advantages and disadvantages of Neural Network-based gas turbine diagnostics have been analysed.

PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO
Turbinas a gás industriais modernas instaladas em usinas termelétricas têm seus parâmetros de desempenho monitorados em tempo real. Contudo, existem inúmeras falhas de operação que são impossíveis de serem detectadas pela simples visualização destes parâmetros, uma vez que a condição de operação do equipamento é influenciada por diversos fatores. Sistemas de diagnóstico são usualmente oferecidos pelos fabricantes destes equipamentos, mas não são divulgados na literatura aberta, que conta em geral com trabalhos aplicados a casos específicos e a turbinas aeronáuticas. Esta dissertação propõe um sistema de diagnóstico de falhas em turbinas a gás, o qual opera através da contínua comparação entre sinais medidos em campo, os quais são simulados por um programa computacional, e resultados gerados por um modelo de referência, simulador da turbina saudável. O sinal comparado serve de entrada para um sistema fuzzy, que identifica e quantifica a severidade das falhas. Foram testadas falhas fictícias no compressor e foi avaliada a influência da mudança de geometria na calibração do sistema. Os resultados mostraram a robustez do sistema e sua capacidade de aplicação em uma situação real.
Modern industrial gas turbines installed in thermal power plants have its performance parameters monitored in real time, however, there are innumerable operation faults that cannot be detected by a simple visual analysis of these parameters, once the equipment operating condition is influenced by several factors. Diagnosis systems are usually offered by the manufacturers of these equipments, but the methodologies are not published in the open literature, which is mostly dedicated to aircraft engines. This dissertation proposes a gas turbine diagnosis system that operates through the continuous comparison between the field measured signals, simulated by a software, and results generated by a reference numerical model, which represents the healthy gas turbine. The compared signal is used as input to a fuzzy system that identifies and quantifies the faults severity. Dummy compressor faults have been tested and the influence of the variable geometry has been analyzed during the system calibration. The results have shown the robustness of the system and its capability to be applied in a real world situation.

Cette thèse est consacrée à l'étude de diagnostic de pannes pour les systèmes pile à combustible de type PEMFC. Le but est d'améliorer la fiabilité et la durabilité de la membrane électrolyte polymère afin de promouvoir la commercialisation de la technologie des piles à combustible. Les approches explorées dans cette thèse sont celles du diagnostic guidé par les données. Les techniques basées sur la reconnaissance de forme sont les plus utilisées. Dans ce travail, les variables considérées sont les tensions des cellules. Les résultats établis dans le cadre de la thèse peuvent être regroupés en trois contributions principales.La première contribution est constituée d'une étude comparative. Plus précisément, plusieurs méthodes sont explorées puis comparées en vue de déterminer une stratégie précise et offrant un coût de calcul optimal.La deuxième contribution concerne le diagnostic online sans connaissance complète des défauts au préalable. Il s'agit d'une technique adaptative qui permet d'appréhender l'apparition de nouveaux types de défauts. Cette technique est fondée sur la méthodologie SSM-SVM et les règles de détection et de localisation ont été améliorées pour répondre au problème du diagnostic en temps réel.La troisième contribution est obtenue à partir méthodologie fondée sur l'utilisation partielle de modèles dynamiques. Le principe de détection et localisation de défauts est fondé sur des techniques d'identification et sur la génération de résidus directement à partir des données d'exploitation.Toutes les stratégies proposées dans le cadre de la thèse ont été testées à travers des données expérimentales et validées sur un système embarqué
Aiming at improving the reliability and durability of Polymer Electrolyte Membrane Fuel Cell (PEMFC) systems and promote the commercialization of fuel cell technologies, this thesis work is dedicated to the fault diagnosis study for PEMFC systems. Data-driven fault diagnosis is the main focus in this thesis. As a main branch of data-driven fault diagnosis, the methods based on pattern classification techniques are firstly studied. Taking individual fuel cell voltages as original diagnosis variables, several representative methodologies are investigated and compared from the perspective of online implementation.Specific to the defects of conventional classification based diagnosis methods, a novel diagnosis strategy is proposed. A new classifier named Sphere-Shaped Multi-class Support Vector Machine (SSM-SVM) and modified diagnostic rules are utilized to realize the novel fault recognition. While an incremental learning method is extended to achieve the online adaptation.Apart from the classification based diagnosis approach, a so-called partial model-based data-driven approach is introduced to handle PEMFC diagnosis in dynamic processes. With the aid of a subspace identification method (SIM), the model-based residual generation is designed directly from the normal and dynamic operating data. Then, fault detection and isolation are further realized by evaluating the generated residuals.The proposed diagnosis strategies have been verified using the experimental data which cover a set of representative faults and different PEMFC stacks. The preliminary online implementation results with an embedded system are also supplied

L'objectif de cette thèse est de construire un système intelligent, efficace et pratique pour diagnostiquer et pronostiquer les pannes d'avions. Mes recherches portent sur "La MOdélisation, le DIagnostic et le PROnostic (MODIPRO) de pannes dans les systèmes complexes". Ce travail s'inscrit dans le cadre d'un projet FUI intitulé MODIPRO, qui est porté par Dassault Aviation, dont l’objectif est de mettre sur marché une solution logicielle. Cette solution logicielle permettra d’analyser la masse de données acquises en vol par un parc d’avions afin d'en déduire des règles de diagnostic et un pronostic de panne. Le système proposé dans cette thèse a été entièrement testé à l'aide de données expérimentales de trois trimoteurs d'avions Z1, Z2 et Z3 (fournies par Dassault Aviation). L'ensemble du système devrait être construit sur une base de données contenant environ 67 heures de vol dossiers impliquant 32 capteurs. Les solutions de diagnostic classiques deviennent de moins en moins applicables pour les avions modernes, dont les systèmes électroniques et mécaniques deviennent de plus en plus complexes. Dans l’état de l’art, la maintenance non planifiée n'a lieu qu'au moment où les pannes surviennent, ce qui est trop tard pour observer les dysfonctionnements; la maintenance planifiée est préventive et doit être réalisée périodiquement et indépendamment de l'état physique de l'avion, ce qui nécessite une quantité importante de ressources financières et humaines. Bien que les tests intégrés soient largement utilisés aujourd’hui, ils prennent également beaucoup du temps car le personnel de maintenance a besoin de se connecter à une boîte de diagnostic de l'appareil après chaque vol. Ces méthodes classiques provoquent souvent un grand nombre de fausses alarmes, par conséquent la maintenance planifiée est encore indispensable aujourd’hui. De plus, les systèmes de diagnostic et de pronostic classique, tels que le management de maintenance conditionnelle (CBM) et la gestion du pronostique de situation (PHM), n’analysent la situation des avions que lorsque ceux-ci sont au sol - en mode «hors-ligne», ils ne peuvent donc pas contrôler des avions en mission. Pour résoudre tous ces problèmes et garantir un taux élevé de participation des aéronefs, le système proposé dans cette thèse utilise des méthodes d'apprentissage automatique pour détecter, isoler et même prévoir les pannes d'avions tout en conservant la fiabilité et la sécurité. Ces recherches font appel aux techniques de traitement du signal, de la reconnaissance des formes et de la classification. D'une part, un modèle de diagnostic permet de déduire la cause « réelle » d’une panne par l’observation et le traitement de signaux acquis en vol. D'autre part, un modèle de pronostic fournit l’état d’avancement d’une dégradation et permet donc d’anticiper ou reporter la maintenance. L’exploitant du système peut utiliser le diagnostic pour identifier et localiser une panne et le pronostic pour arbitrer entre ses besoins d’exploitation, ses coûts de remise en état, les risques de défaillance et leurs conséquences. En plus, ce système peut être utilisé non seulement en mode «hors-ligne», lors de la maintenance d'un aéronef, mais aussi en mode «en ligne», lorsque celui-ci est en mission. Selon les exigences de la situation, les missions du système en ligne et du système hors ligne peuvent être différentes. Le système en ligne est chargé de détecter les pannes et d’envoyer des alarmes au pilote et à la tour de contrôle. Le système hors-ligne nécessite de localiser les pannes et de faire un rapport détaillé au centre de maintenance. En outre, le système a besoin d'analyser les données des vols effectués afin de prévoir des pannes. Afin de garantir la fiabilité du système, différentes méthodes d’apprentissage sont connectées en parallèle comme des sous-systèmes. Ces méthodes peuvent compenser les inconvénients de l'autre. Dans un premier temps, les données sont analysées et pré-classées par une approche classique et simple - l'analyse discriminante linéaire (LDA). Sur la base de ces résultats, une nouvelle approche de la classification appelée SCM est proposé d'améliorer la précision du diagnostic. SCM est différent de SVM qui exige des vecteurs de support à la frontière de chaque classe pour distinguer les différentes catégories. SCM cherche les vecteurs de support de centres et sous-centres véritables de chaque classe au cours de l'apprentissage automatique. Il peut utiliser les centres correspondants comme le modèle de la classe. La classification des données se fait simplement d'après l'éloignement des centres. En outre, SCM peut travailler pour l'analyse pronostique et parfaitement résoudre le problème même dans le cas où les données sont non linéaires. L'évolution temporelle des données de vol est analysée par chaque modèle de panne. Sur la base de l'évolution de la distance entre le nuage de données et les centres du modèle, le système calcule la tendance de l'évolution des données et prévoit les pannes probables. Au-delà d'un pronostic de panne à court terme, le système peut également être utilisé pour faire une évaluation à long terme de l'état des aéronefs. Ceci est plus convaincant et efficace par rapport aux méthodes de régression et aux méthodes statistiques, qui n'ont pas la précision d'une régression à long terme et qui nécessitent plus de temps pour l'analyse des données. Bien que les résultats de diagnostic des SCM et SVM soient déjà satisfait, avec un taux de détection correcte qui dépasse 95%, des réseaux de neurones artificiels (ANN) sont utilisés pour construire un autre sous-système, afin d’analyser l’impact des différents types des capteurs et confirmer les modèles produits par SCM et SVM. Les ANN sont une approche tout à fait différente de SCM et SVM : il s'agit d'un modèle mathématique qui est inspiré par les aspects de structure et fonctionnelles des réseaux de neurones biologiques. Un réseau de neurones est constitué d'un groupe interconnecté de neurones artificiels, et il traite les informations en utilisant une approche connexionniste de calcul. Les capteurs sont répartis en différents groupes correspondants à leur type : température, pression de l'air, etc. Ces groupes des capteurs constituent les entrées des réseaux de neurones, ainsi nous pouvons étudier l'importance de chaque type de capteur d'après leurs poids dans le réseau et les résultats du diagnostic des pannes. Avec ces résultats, on peut déterminer quels groupes des capteurs sont les plus importants pour diagnostiquer chaque type de panne. Les méthodes SCM, SVM et ANN ont besoin beaucoup de temps pour réaliser l’apprentissage, ce qui ne permet pas d’apprendre au cours des vols. Dans certains cas, il peut être nécessaire de reconstruire le système de diagnostic et de pronostic, par exemple si un capteur est perdu pendant la mission. Pour pallier à cela, nous avons ajouté des sous-systèmes basés sur des arbres de décision (DT) et des modèles des mélanges Gaussiens (GMM). L'algorithme C4. 5 apprend automatiquement le meilleur arbre de décision en effectuant une recherche dans l'ensemble des arbres possibles selon les données d'apprentissage disponibles. Donc il est capable de travailler même avec des éléments d'information manquants. Il peut être utilisé pour construire un sous-système capable de restructurer le système de diagnostic à temps si certains capteurs ou informations sont perdus. Les GMM permettent de dessiner le plan des modèles dysfonctionnelles pour surveiller l’évolution des données réelles de l'avion dans le système de pronostic. Contrairement aux systèmes d’experts ou à d'autres méthodes classiques, les méthodes développées dans cette thèse peuvent facilement intégrer de nouvelles pannes et de nouvelles règles dans la base de données : il n'y a pas de conflit entre les nouvelles et les anciennes règles. D’autre part, les capteurs sont susceptibles de tomber en panne, certaines entrées peuvent donc manquer au système. Les mesures des capteurs étant utilisées comme entrées du système, la nature des capteurs influe sur l'exactitude des résultats de diagnostic et de pronostic, ainsi que dans la confiance que l'on peut avoir dans ceux-ci. Pour traiter ces problèmes, le système doit constamment vérifier l’état des capteurs à l’aide d’un modèle physique. Si certains capteurs sont en panne, le système d'origine n'est pas applicable. Il faut alors démarrer une solution d'urgence, comme le réapprentissage rapide de l'arbre de décision afin de construire un nouveau système de diagnostic de panne temporaire. En plus de cela, l'analyse en composantes principales (PCA) et l'analyse discriminante linéaire (LDA) peuvent non seulement réduire la dimension des données d'entrée, mais permettent aussi de visualiser les données en 2D ou en 3D. Ces outils sont très utiles pour observer le pronostic de panne d’un avion ou d’une flotte d’avions à partir des données réelles. D’autre part, ces informations peuvent servir aux ingénieurs pour étudier la nature des pannes observées. Le système décrit ici n'est pas une boîte noire. Bien qu'il soit construit principalement pour les avions de combat, il peut être aussi appliqué à tous les autres types d'avions, nommément des avions civile. D'une part, le système et ses modèles détectant les dysfonctionnements potentiels peuvent être conçus pour éclairer les services de client chargés de surveiller l'état des avions afin d'assurer la sécurité des clients. D'autre part, ce système peut également accumuler les connaissances (y compris les règles de fonctionnement) pour le bureau d’études et parfaire la conception de nouveaux avions
The goal of this thesis is to build an effective and practical intelligent system to diagnose and prognose aircraft faults. My research focuses on “The MOdeling, DIagnosis and PROgnosis (MODIPRO)” faults in complex systems. This work is a part of a project entitled FUI MODIPRO which is supported by Dassault Aviation. The objective of this project is to research and develop a software solution MODIPRO Version 0 and put it on the aviation market. This software solution can analyze a huge mass of data acquired from a flight and a fleet of aircraft, and the system can deduce rules for diagnosis and prognosis of faults. The system proposed in this thesis has been fully tested by using actual experimental data from a tri-engines system of aircrafts Z1, Z2 and Z3 (supplied by Dassault Aviation). The whole system would be built on a database containing about 67 hours of flight records involving 32 sensors. With the rapid development of modern aero technology and the market demand of high- performance, aircraft systems have become more and more. Thus, the classical diagnosis methods become less available. In the state of the art, unplanned maintenance takes place only at breakdowns, which is too late to observe the faults; the planned maintenance costs too much financial resources and manpower, which needs to set a periodic interval to perform preventive maintenance regardless of the health status of a physical asset. Although Build in Test (BIT) system is used widely, it also costs too much human and financial resource. In a general way the maintenance staffs need to connect the diagnostic box to the aircraft via interface after each flight mission. Because these classical methods often cause the false alarm, the planned maintenance is also indispensable today. In addition, classical diagnostic and prognostic system, such as Condition-Based Maintenance (CBM) and Prognostic Health Management (PHM), analyze the health state of aircrafts when they are on the ground – in the "offline" mode, they can’t supervise the aircraft during the mission. In order to resolve these problems and guarantee a high ratio of attendance of aircraft, the system proposed in this thesis uses machine-learning methods to automatically detect, isolate, and even forecast aircraft faults while maintaining reliability and safety. The researches involve signals processing techniques, pattern recognition and classification. On the one hand, the diagnostic model allows the system to deduce the "real" cause of a fault by the observation and the treatment of acquired signals from flight records. On the other hand, the model can provide a progress of degradation of the health state and thus allows anticipating the faults or deferring the needless planned maintenance. The diagnosis system can locate and identify faults and the prognosis system can make the arbitration of a future maintenance plan on basis of the operating needs, the costs of rehabilitation, the risk of fault and the consequences. In addition to this, the system proposed in this thesis can be used not only in the off-line mode when aircraft maintenance occurs, but also in the on-line mode during the aircraft’s mission. According to the different situations requirements, the missions of on-line system and off-line system are different. The on-line system is tasked with detecting faults and sending the alarms to the pilot and the Aircraft Ground Center (AGC) in time. The off-line system is obliged to locate the fault(s) and make a detail report to the maintenance center. Additionally, the system needs to analyze the flight data in the past time for the sake of forecasting the fault(s). In order to ensure the reliability of the system, different methods of machine learning are used in parallel as subsystems. These methods can compensate the disadvantages of each other. At first, the data are analyzed and pre-classified by Linear Analysis Discriminant (LDA), a classical and simple approach. On basis of the results, a novel approach of classification called SCM is proposed to improve the accuracy of diagnosis. SCM is different from SVM that requires the support vectors on the boundary of every class to distinguish the categories. SCM seeks the support vectors of true centers and sub-centers of each class during the machine learning. It can make the corresponding centers as the model of the class. The classification of data is simply done by the power distances of the centers. Furthermore, SCM can work for the prognosis analysis and perfectly deal with the nonlinear problem. The evolution of flight data is supervised by each fault model. On the basis of the evolution of the distances from the cloud of data to the centers, the system estimates the tendency of the evolution of data and forecast the probable faults in the future. Beyond a short-term prognosis of faults, the system can also be used to do a long-term evaluation of aircraft healthy state. This is more convincing and efficacious compared to regression methods and statistical methods, which lack the precision of a long-term regression and which require a longer time for data analysis. Although the diagnosis results of SCM and SVM are already satisfied with a correct detection rate that exceeds 95%, Artificial Neural Networks (ANN) are used to build another sub-system, so as to analyze the impact of using different types sensors on the different fault diagnosis and confirm the results from the models SVM and SCM. ANN is a quite different AI technic from SCM and SVM. It is a mathematical model that is inspired by the structure and functional aspects of biological neural networks. A neural network consists of an interconnected group of artificial neurons, and it processes information using a connectionist approach to computation. All the sensors are divided in to different groups corresponding to different types of the sensors. Different combinations of sensors are linked to the neural networks, thus we can study the importance of different types of aircraft sensors by the weights of networks and the diagnosis results of the faults. The methods, as SCM, SVM and ANN, need much time to accomplish machine learning, which cannot do the learning during the flight mission. But, in some cases, it may be necessary to rebuild the diagnosis system, for example if some sensors are broken or lost during the mission. For overcoming this, we added sub-systems based on decision trees (DT) and Gaussian mixture models (GMM), which are easier to interpret, quicker to learn than other data-driven methods, and able to work even with missing pieces of information. The C4. 5 algorithm automatically "learns" the best decision tree by performing a search through the set of possible trees according to the available training data. Its needs less time to accomplish the machine learning, so it is also studied and improved in this thesis, and be used to build a subsystem for sake of restructuring the diagnosis system if some sensors or sensors information are lost, especially under the condition of war. GMM can also draw the plan of dysfunctional models and monitor the evolution of the health state of the aircraft in the prognosis system. Unlike expert systems or other conventional methods, the methods developed in this thesis can easily integrate new faults and new rules in the database: there isn’t any conflict between the new and old rules. Beyond that, there is another important problem to consider and resolve: some sensors might be already failed before the machine learning. The measurements via sensors in the aircraft are used as the inputs of the system. The nature of the sensors will impact the accuracy and confidence of the diagnosis and prognosis results of the system. Thus, these data should be treated above of all. First, the system needs to check the healthy state of the sensors. If some sensors are broken down, the original system is not applicable. The system will start the emergency application, like fast relearning of the decision tree in order to build a new temporary fault diagnosis system. In addition to that, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used in data mining. They can not only reduce the input data’s dimension, but also make a visualization of data in 2D or 3D. It is very useful to observe the evaluation of flux data and to realize prognosis, and it is important for engineers to study the nature of faults. The system described here is not a black box. Although the system is built mainly for combat aircraft, it can be applied to all other types of aircraft, namely civil aircraft. On one hand, the system and its dysfunction models of aircraft faults can be designed to illuminate engineering consulting services responsible for monitoring the condition of aircrafts to ensure the safety of clients. On the other hand, this system can also accumulate the knowledge for re-engineering purposes (including diagnosis operational rules) and perfect the design of new aircrafts

The implementation of an automatic fault diagnosis approach for analogue circuits is facing a number of problem areas. They are typically: component and measurement tolerances, circuit size, limited observability constraints, multiple fault conditions, non-linear behaviour, speed and generic applicability. Since such fault finding techniques utilize circuit simulations sometime during the diagnostic process, the preferred form of classification amongst researchers is a taxonomy of Simulationbefore-Test (SbT) and Simulation-after-Test (SaT) methods. A survey of related work following these two strategies has been carried out, which concludes: The main advantage of the SaT strategy is their diagnostic power to cope with above problem areas, their main disadvantage is the often considerable computational on-line effort. The main advantage of the SbT strategy is on-line speed, but diagnostic power is often limited. What is needed is a workable solution to combine the advantages of the two strategies, whilst minimizing their disadvantages. The thesis is focused on this need. Subject of the research programme was therefore to look into the feasibility of a Simulation-before-Test approach for diagnosing toleranced analogue non-linear networks in the presence of multiple faults and from there to research the concepts, strategies and algorithms required to form a diagnostic approach.

The theme of cooling is recurrent in the world around us. Air conditioning systems in residential environments are the most common examples of refrigeration systems. However, also in the treatment, storage, transport and distribution of food products, as well as in the health and tertiary sectors, refrigeration plays a central role. The main purpose of this research is the analysis of some techniques for the detection and diagnosis of faults in this type of systems, also called chillers. In the work, the chiller is analyzed in all its components, for which the operating principle and the significant variables for the fault detection task are derived. The research proceeds to the analysis of static methodologies based on the data for the detection of anomalies. Each of them builds a model of the system. This model is then used in the monitoring stage. The static nature of the methods proposed in the thesis refers to the use, in the model identification phase, of data relating to steady state of the system instead of the entire time evolution of the signals. In this way, the system is monitored in conditions of thermodynamic stationarity and sudden transients, difficult to characterize mathematically, are eliminated from the final database. The choice of data-driven methods is consistent with the direction of the current literature, mainly focused on those approaches that do not require a detailed physical description of the system. The ability to fine-tune the model from the data makes these techniques easily applicable to different plants. In particular, the thesis considers three techniques for the detection of anomalies. Two of them, the multiple linear regression and the Principal Components Analysis (PCA), identify a model for the data in the form, respectively, of a surface and a regression hyper-plane, while the third, the Mahalanobis's distance, takes into account the probabilistic characteristics of the dataset. These techniques are generally used for the prediction or for the dimensional reduction. In the thesis their effectiveness is tested in the context of the detection of anomalies. The different philosophies from which they take inspiration and the advantages and disadvantages of each approach are considered. The comparison is proposed for some faulty dataset generated with software and on a real case.
Il tema del raffrescamento è ricorrente nel mondo che ci circonda: i sistemi di climatizzazione negli ambienti residenziali sono gli esempi più comuni di sistemi di refrigerazione. Tuttavia anche nel trattamento, stoccaggio, trasporto e distribuzione di prodotti alimentari, così come nel settore sanitario e terziario, la refrigerazione svolge un ruolo centrale. Lo scopo principale della ricerca è l'analisi di alcune tecniche per l'individuazione e la diagnosi di guasti in questa tipologia di sistemi, anche detti chillers. All'interno del lavoro, il chiller è analizzato in tutti i suoi componenti, per i quali vengono dedotti il principio di funzionamento e le variabili significative per la rilevazione dei guasti. La ricerca procede all'analisi di metodologie statiche basate sui dati per il rilevamento di anomalie. Ognuna di esse prevede la costruzione di un modello del sistema; tale rappresentazione viene poi utilizzata nella fase di monitoraggio. La natura statica dei metodi proposti nella tesi riferisce all'uso, nella fase di identificazione del modello, di dati relativi a stati stazionari del sistema invece dell'intera evoluzione temporale dei segnali. In questo modo, il sistema è monitorato in condizioni di stazionarietà termodinamica e transitori improvvisi, difficili da caratterizzare matematicamente, sono eliminati dal database finale. La scelta di metodi basati sui dati è coerente con la direzione della letteratura corrente focalizzata su quegli approcci che non richiedono una descrizione fisica dettagliata del sistema monitorato. La possibilità di mettere a punto il modello dai dati rende tali tecniche facilmente applicabili a differenti impianti. In particolare, la tesi considera tre tecniche per la rilevazione di anomalie. Due di esse, la regressione lineare multipla e l'Analisi delle Componenti Principali (PCA), identificano un modello per i dati nella forma, rispettivamente, di una superficie e di un iperpiano di regressione, mentre la terza, la distanza di Mahalanobis, prende in considerazione le caratteristiche probabilistiche dell'insieme di dati. Queste tecniche sono generalmente utilizzate a scopo previsionale o per la riduzione dimensionale: nella tesi ne viene testata l'efficacia nel contesto della rilevazione di anomalie, illustrando le diverse filosofie dalle quali esse prendono spunto e commisurandone vantaggi e svantaggi. Il confronto viene proposto per degli insiemi di guasti simulati via software e per un caso reale.

Wind energy is the fastest-growing energy source in the world nowadays and most wind turbines are installed at remote areas, e.g. country side, off sea-shore. Having a reliable fault diagnosis and fault tolerant control (FTC) scheme is crucial to improve the reliability of wind turbines and reduce expensive repair cost. This PhD work is motivated by this fact and a model-based fault diagnosis and FTC scheme is developed for a doubly fed induction generator (DFIG) based wind turbine system. In particular, an electrical and a mechanical fault scenarios, the DFIG winding short circuit and drive train faults, are considered due to their high occurrence rates.For the DFIG winding short circuit fault, two mathematical models of DFIG with respect to two types of faults, i.e. single-phase and multi-phase faults, are proposed which can represent all possible cases of the faults. Moreover, the state-space representations of these models are derived by using reference frame transformation theory, such that the faults are represented by some unknown variables or parameters. Based on these models, an adaptive observer based fault diagnosis scheme is proposed to diagnose short circuit faults via online estimation of unknown variables or parameters. By dong this, the fault level and location can be online diagnosed. To consider the effects of model uncertainties, two robust adaptive observers are proposed based on the H∞ optimization and high-gain observer techniques, respectively, which can ensure the accuracy and robustness of fault estimations. In addition, a self-scheduled LPV adaptive observer is developed with consideration of rotor speed variations, which is suitable for the fault diagnosis under non-stationary conditions. In the context of FTC, a fault compensator is developed based on fault information provided by the fault diagnosis scheme, and it incorporates with a traditional controller (i.e. stator flux oriented controller) to provide an online fault compensation of winding short circuit faults.For the mechanical drive train fault, the work focuses on FTC rather than diagnosis. Without using an explicit fault diagnosis scheme, an active FTC scheme is directly designed by employing an adaptive input-output linearizing control (AIOLC) technique. It provides a perfect reference tracking of the torque and reactive power no matter whether the fault occurs. In addition, a robust AIOLC is proposed in order to ensure FTC performance against model uncertainties.

The selection of adequate fault models is crucial to generating tests of high quality for complex digital VLSI circuits. This thesis presents a methodology to perform empirical validation of fault models and to get measures of effectiveness of test sets based on the targeted fault models.
The methodology is based on the automated fault diagnosis of test circuits, representative of the class of circuits being studied and designed to capture the characteristics of the fabrication process, cell libraries and CAD tools used in their development.
The methodology is applied to study the faulty behaviour of random logic environments for an experimental VLSI fabrication process. A test circuit is designed, using CMOS technology, and a statistically significant number of samples fabricated. The samples are tested and, subsequently, diagnosed, using a set of software tools developed for the purpose. Results of the ensuing analysis are presented.

The aim of this study is to solve Fault Detection and Diagnosis (FDD) problems occurring in nonlinear dynamical systems by using model and knowledge-based FDD methods and to give a priority and a degree about faults. For this purpose, three model-based FDD approaches, called FDD by utilizing principal component analysis (PCA), system identification based FDD and inverse model based FDD are introduced. Performances of these approaches are tested on different nonlinear dynamical systems starting from simple to more complex. New fuzzy discrete event system (FDES) and fuzzy discrete event dynamical system (FDEDS) concepts are introduced and their applicability to an FDD problem is investigated. Two knowledge-based FDD methods based on FDES and FDEDS structures using a fuzzy rule-base are introduced and they are tested on nonlinear dynamical systems. New properties related to FDES and FDEDS such as fuzzy observability and diagnosibility concepts and a relation between them are illustrated. A dynamical rule-base extraction method with classification techniques and a dynamical and a static diagnoser design methods are also introduced. A nonlinear and event based extension of the Luenberger observer and its application as a diagnoser to isolate faults are illustrated. Finally, comparisons between the proposed model and knowledge-based FDD methods are made.