Thesis (PhD (Process Engineering))--University of Stellenbosch, 2010. / Dissertation presented for the Degree of DOCTOR OF PHILOSOPHY (Extractive Metallurgical Engineering) in the Department of Process Engineering at the University of Stellenbosch / ENGLISH ABSTRACT: Fault diagnosis is an important component of process monitoring, relevant in the greater context of developing safer, cleaner and more cost efficient processes. Data-driven unsupervised (or feature extractive) approaches to fault diagnosis exploit the many measurements available on modern plants. Certain current unsupervised approaches are hampered by their linearity assumptions, motivating the investigation of nonlinear methods. The diversity of data structures also motivates the investigation of novel feature extraction methodologies in process monitoring. Random forests are recently proposed statistical inference tools, deriving their predictive accuracy from the nonlinear nature of their constituent decision tree members and the power of ensembles. Random forest committees provide more than just predictions; model information on data proximities can be exploited to provide random forest features. Variable importance measures show which variables are closely associated with a chosen response variable, while partial dependencies indicate the relation of important variables to said response variable. The purpose of this study was therefore to investigate the feasibility of a new unsupervised method based on random forests as a potentially viable contender in the process monitoring statistical tool family. The hypothesis investigated was that unsupervised process monitoring and fault diagnosis can be improved by using features extracted from data with random forests, with further interpretation of fault conditions aided by random forest tools. The experimental results presented in this work support this hypothesis. An initial study was performed to assess the quality of random forest features. Random forest features were shown to be generally difficult to interpret in terms of geometry present in the original variable space. Random forest mapping and demapping models were shown to be very accurate on training data, and to extrapolate weakly to unseen data that do not fall within regions populated by training data. Random forest feature extraction was applied to unsupervised fault diagnosis for process data, and compared to linear and nonlinear methods. Random forest results were comparable to existing techniques, with the majority of random forest detections due to variable reconstruction errors. Further investigation revealed that the residual detection success of random forests originates from the constrained responses and poor generalization artifacts of decision trees. Random forest variable importance measures and partial dependencies were incorporated in a visualization tool to allow for the interpretation of fault conditions. A dynamic change point detection application with random forests proved more successful than an existing principal component analysis-based approach, with the success of the random forest method again residing in reconstruction errors. The addition of random forest fault diagnosis and change point detection algorithms to a suite of abnormal event detection techniques is recommended. The distance-to-model diagnostic based on random forest mapping and demapping proved successful in this work, and the theoretical understanding gained supports the application of this method to further data sets. / AFRIKAANSE OPSOMMING: Foutdiagnose is ’n belangrike komponent van prosesmonitering, en is relevant binne die groter konteks van die ontwikkeling van veiliger, skoner en meer koste-effektiewe prosesse. Data-gedrewe toesigvrye of kenmerkekstraksie-benaderings tot foutdiagnose benut die vele metings wat op moderne prosesaanlegte beskikbaar is. Party van die huidige toesigvrye benaderings word deur aannames rakende liniariteit belemmer, wat as motivering dien om nie-liniêre metodes te ondersoek. Die diversiteit van datastrukture is ook verdere motivering vir ondersoek na nuwe kenmerkekstraksiemetodes in prosesmonitering. Lukrake-woude is ’n nuwe statistiese inferensie-tegniek, waarvan die akkuraatheid toegeskryf kan word aan die nie-liniêre aard van besluitnemingsboomlede en die bekwaamheid van ensembles. Lukrake-woudkomitees verskaf meer as net voorspellings; modelinligting oor datapuntnabyheid kan benut word om lukrakewoudkenmerke te verskaf. Metingbelangrikheidsaanduiers wys watter metings in ’n noue verhouding met ’n gekose uitsetveranderlike verkeer, terwyl parsiële afhanklikhede aandui wat die verhouding van ’n belangrike meting tot die gekose uitsetveranderlike is. Die doel van hierdie studie was dus om die uitvoerbaarheid van ’n nuwe toesigvrye metode vir prosesmonitering gebaseer op lukrake-woude te ondersoek. Die ondersoekte hipotese lui: toesigvrye prosesmonitering en foutdiagnose kan verbeter word deur kenmerke te gebruik wat met lukrake-woude geëkstraheer is, waar die verdere interpretasie van foutkondisies deur addisionele lukrake-woude-tegnieke bygestaan word. Eksperimentele resultate wat in hierdie werkstuk voorgelê is, ondersteun hierdie hipotese. ’n Intreestudie is gedoen om die gehalte van lukrake-woudkenmerke te assesseer. Daar is bevind dat dit moeilik is om lukrake-woudkenmerke in terme van die geometrie van die oorspronklike metingspasie te interpreteer. Verder is daar bevind dat lukrake-woudkartering en -dekartering baie akkuraat is vir opleidingsdata, maar dat dit swak ekstrapolasie-eienskappe toon vir ongesiene data wat in gebiede buite dié van die opleidingsdata val. Lukrake-woudkenmerkekstraksie is in toesigvrye-foutdiagnose vir gestadigde-toestandprosesse toegepas, en is met liniêre en nie-liniêre metodes vergelyk. Resultate met lukrake-woude is vergelykbaar met dié van bestaande metodes, en die meerderheid lukrake-woudopsporings is aan metingrekonstruksiefoute toe te skryf. Verdere ondersoek het getoon dat die sukses van res-opsporing op die beperkte uitsetwaardes en swak veralgemenende eienskappe van besluitnemingsbome berus. Lukrake-woude-metingbelangrikheidsaanduiers en parsiële afhanklikhede is ingelyf in ’n visualiseringstegniek wat vir die interpretasie van foutkondisies voorsiening maak. ’n Dinamiese aanwending van veranderingspuntopsporing met lukrake-woude is as meer suksesvol bewys as ’n bestaande metode gebaseer op hoofkomponentanalise. Die sukses van die lukrake-woudmetode is weereens aan rekonstruksie-reswaardes toe te skryf. ’n Voorstel wat na aanleiding van hierde studie gemaak is, is dat die lukrake-woudveranderingspunt- en foutopsporingsmetodes by ’n soortgelyke stel metodes gevoeg kan word. Daar is in hierdie werk bevind dat die afstand-vanaf-modeldiagnostiek gebaseer op lukrake-woudkartering en -dekartering suksesvol is vir foutopsporing. Die teoretiese begrippe wat ontsluier is, ondersteun die toepassing van hierdie metodes op verdere datastelle.

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 have become an indispensable tool within most industrial sectors and are used to reduce costs, enhance the performance of a system etc. The computer support within modelling is extensive, whereof the programming language Modelica is eminent, especially for multi-domain models. Dymola, a commercial program, is built on Modelica and is foremost used for simulation purposes, but many applications for which models are useful are not supported by Dymola. Instead other tools, e.g. Matlab, could be used to exploit the full potential of a model, which means that it first would be needed to be translated. This master's thesis examines one of the possible ways to accomplish this. Specifically the possibility to translate Modelica-models via an XML file, generated by Dymola, is examined. The structure and content of this file is explored, and based thereupon a software is implemented in Python, which successfully translates the models constituting the base for this thesis. Specifically the method was developed on a model of a sub-system of Saab 39 Gripen air-plane. Besides porting models between different languages, it is of great interest to determine how well a model describes the system on which it is based. Hence a new method for model validation is developed using the Matlab Fault Diagnosis Toolbox, which also determines the Matlab syntax of the Modelica translation. The novelty with the developed method, compared to traditional model validation methods, is that it is equation based. It is meant to point out specifically which equations are poorly fitted to validation data. On a simple example model the method was successfully used to isolate a poorly fitted equation. This is accomplished by introducing faults to the equations and generating residuals, based on sets of over-determined equations. As a measure of the modelling error the estimation error of the simulated residuals is used, which are weighted together depending on the fault properties of the residuals.

Dans cette thèse, la conception d’observateurs pour les systèmes singuliers linéaires incertains et leurs applications au contrôle et au diagnostic. En effet, nous avons développé des méthodes de reconstruction d’état et d’estimation de défauts est étudié. Les systèmes algèbro-différentiels ou systèmes singuliers peuvent être considérés comme une généralisation des systèmes dynamiques. Ils constituent un puissant outil de modélisation dans la mesure où ils peuvent décrire des processus régis à la fois par des équations différentielles (dynamiques) et des équations algébriques (statiques). La nouvelle structure d’observateurs utilisée dans cette thèse est nommée l’Observateur Dynamique Généralisé (ODG), elle est plus générale que celle d’Observateurs Proportionnels (OP) et d’Observateurs Proportionnels Intégrals (OPI). Cette structure présente une estimation d’état alternative qui peut être considérée comme plus générale que les OP et les OPI, ceux-ci pouvant être considérés comme des cas particuliers de cette structure. L’approche proposée repose sur la paramétrisation des solutions des équations de Sylvester pour éliminer le biais entre l’erreur d’observation et la paire (entrée état). La thèse est organisée comme suit : Dans l’introduction générale, nous présentons la problématique et les objectifs de la thèse ainsi que les principales contributions. Dans le premier chapitre, nous présentons la classe des systèmes singuliers considérée. Nous faisons des rappels sur l’analyse de stabilité et l’utilisation des outils numériques LMI avec lesquels nous vérifions l’existence de conditions de stabilité. Ensuite, nous présentons les méthodes de reconstruction d’état des systèmes singuliers linéaires à savoir l’ODG, l’OP et l’OPI. Dans le deuxième chapitre, nous présentons en détail la procédure de synthèse d’ODG pour les systèmes singuliers continus avec et sans perturbations. Ensuite, nous faisons une extension aux systèmes singuliers en temps discret avec et sans perturbations. Dans le chapitre 3, nous donnons les conditions d’existence et de stabilité robuste de l’ODG pour les systèmes singuliers à paramètres incertains, où l’incertitude est bornée. Dans le chapitre 4, nous présentons une méthode de synthèse de commande stabilisante par retour d’état basée observateur pour une classe de systèmes singuliers linéaires avec et sans perturbations. Le chapitre 5, est consacré au diagnostic. L’étude que nous avons menée est traitée en deux étapes : La première étape est consacrée à la détection et l’isolation des défauts en utilisant un ODG. Cet observateur génère des résidus qui sont en mesure de représenter seulement la présence d’un défaut, de sorte que nous pouvons localiser des défauts multiples. Enfin, la deuxième étape est consacrée à l’estimation des défauts en utilisant un ODG avec une structure modifiée. Ces approches sont développées pour les systèmes singuliers et pour les systèmes singuliers incertains avec ou sans perturbations. Nous terminerons ce mémoire de thèse par une conclusion générale et quelques perspectives. / In this thesis the observer design for uncertain linear descriptor systems and their applications to control and fault diagnosis is studied. Descriptor systems can be considered as a generalization of dynamical systems. This class of systems include algebraic and differential equations. The observer used in this work has a new structure more general than those presented in the literature. The observer structure proposed has additional degrees of freedom, which provides it robustness in face to variations not considered in the model. The new observer structure used in this thesis, named as generalized dynamic observer (GDO), is designed for different classes of descriptor systems. The asymptotic stability of the observer is proved by Lyapunov analysis through a set of linear matrix inequalities (LMIs). In all cases, the LMI obtained from the Lyapunov analysis is treated by the elimination lemma. The use of the elimination lemma is essential in the development of the stability analysis of the observers, since it allows to obtain the GDO structure. Proportional observers (PO) and proportional-integral observers (PIO) can be considered as particular cases of our observer. The thesis is organized as follows: In the general introduction, the problem formulation is presented, the objectives of the thesis are pointed out, the scope of the investigation and the main contributions are also presented. Chapter 1 introduces descriptor systems as the class of systems considered in this work and presents a review of the state of the art focused on the observers design for these systems. Also we introduce the GDO as an observer with structure more general than that of the PO and the PIO. Chapter 2 develops the GDO for descriptor systems with or without disturbances. Extension of these approaches for discrete-time descriptor systems with or without disturbances are also presented. In Chapter 3, the robust approach of the GDO is treated for parametric uncertain descriptor systems, where the uncertainty is bounded, and for linear parameter varying (LPV) descriptor systems, where the parameters vary inside a polytope. Chapter 4 presents the GDO application to observer-based control with the objective to stabilize descriptor systems that normally are unstable. An extension of this approach to disturbed descriptor systems is also developed. Chapter 5 presents the GDO application to fault diagnosis, which is divided in two parts. The first one is to detect and isolate faults by using a GDO that provides residuals that are able to represent only the presence of one fault, so that we can isolate multiple faults. And the second part is to estimate the faults by using a GDO with a modified structure. These approaches are developed for descriptor systems and for uncertain descriptor systems. The last part is dedicated to general conclusions and some perspectives.

Field-Programmable Gate Arrays (FPGAs) são largamente utilizadas em sistemas digitais por características como flexibilidade, baixo custo e alta densidade. Estas características advém do uso de células de SRAM na memória de configuração, o que torna estes dispositivos suscetíveis a erros induzidos por radiação, tais como SEUs. TMR é o método de mitigação mais utilizado, no entanto, possui um elevado custo tanto em área como em energia, restringindo seu uso em aplicações de baixo custo e/ou baixo consumo. Como alternativa a TMR, propõe-se utilizar DMR associado a um mecanismo de reparo da memória de configuração da FPGA chamado scrubbing. O reparo de FPGAs em sistemas em tempo real apresenta desafios específicos. Além da garantia da computação correta dos dados, esta computação deve se dar completamente dentro do tempo disponível (time-slot), devendo ser finalizada antes do tempo limite (deadline). A diferença entre o tempo de computação dos dados e a deadline é chamado de slack e é o tempo disponível para reparo do sistema. Este trabalho faz uso de scrubbing deslocado dinâmico, que busca maximizar a probabilidade de reparo da memória de configuração de FPGAs dentro do slack disponível, baseado em um diagnóstico do erro. O scrubbing deslocado já foi utilizado com técnicas de diagnóstico de grão fino (NAZAR, 2015). Este trabalho propõe o uso de técnicas de diagnóstico de grão grosso para o scrubbing deslocado, evitando as penalidades de desempenho e custos em área associados a técnicas de grão fino. Circuitos do conjunto MCNC foram protegidos com as técnicas propostas e submetidos a seções de injeção de erros (NAZAR; CARRO, 2012a). Os dados obtidos foram analisados e foram calculadas as melhores posição iniciais do scrubbing para cada um dos circuitos. Calculou-se a taxa de Failure-in-Time (FIT) para comparação entre as diferentes técnicas de diagnóstico propostas. Os resultados obtidos confirmaram a hipótese inicial deste trabalho que a redução do número de bits sensíveis e uma baixa degradação do período do ciclo de relógio permitiram reduzir a taxa de FIT quando comparadas com técnicas de grão fino. Por fim, uma comparação entre as três técnicas propostas é feita, analisando o desempenho e custos em área associados a cada uma. / Field-Programmable Gate Arrays (FPGAs) are widely used in digital systems due to characteristics such as flexibility, low cost and high density. These characteristics are due to the use of SRAM memory cells in the configuration memory, which make these devices susceptible to radiation-induced errors, such as SEUs. TMR is the most used mitigation technique, but it has an elevated cost both in area as well as in energy, restricting its use in low cost/low energy applications. As an alternative to TMR, we propose the use of DMR associated with a repair mechanism of the FPGA configuration memory called scrubbing. The repair of FPGA in real-time systems present a specific set of challenges. Besides guaranteeing the correct computation of data, this computation must be completely carried out within the available time (time-slot), being finalized before a time limit (deadline). The difference between the computation time and the deadline is called the slack and is the time available to repair the system. This work uses a dynamic shifted scrubbing that aims to maximize the repair probability of the configuration memory of the FPGA within the available slack based on error diagnostic. The shifted scrubbing was already proposed with fine-grained diagnostic techniques (NAZAR, 2015). This work proposes the use of coarse-grained diagnostic technique as a way to avoid the performance penalties and area costs associated to fine-grained techniques. Circuits of the MCNC suite were protected by the proposed techniques and subject to error-injection campaigns (NAZAR; CARRO, 2012a). The obtained data was analyzed and the best scrubbing starting positions for each circuit were calculated. The Failure-in-Time (FIT) rates were calculated to compare the different proposed diagnostic techniques. The obtained results validated the initial hypothesis of this work that the reduction of the number of sensitive bits and a low degradation of the clock cycle allowed a reduced FIT rate when compared with fine-grained diagnostic techniques. Finally, a comparison is made between the proposed techniques, considering performance and area costs associated to each one.

Power system faults normally result in system damage, profit loss and consumer dissatisfaction. Consequently, there is a strong demand on precise and fast fault location estimation for power system to minimize the system restoration time. This paper examines a method to locate short-circuit faults on a distribution system with unbalanced loads without fault type information. Bus impedance matrix technique was harnessed in the fault location estimation algorithm. The system data including line impedances, source impedance and distribution system layout was assumed to be known factors, hence pre-fault bus impedance can be calculated and implemented into the algorithm. Corresponding methods to derive system matrix information were discussed. Case studies were performed to evaluate the accuracy of the fault location algorithm and illustrate the robust performance under measurements errors influences, load variation impacts and load compensation implementations. Traditional fault location methods involve current and voltage measurements mandatorily locating at each ends of faulted section to locate the fault. The method examined finds fault location for distribution system utilizing impedance matrix accompanied with sparse measurements in the power network. This method fully considers the unbalance of distribution system.

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.

Les travaux de recherche traités dans cette thèse s’appuient sur l’expertise des actionsmenées entre l’Agence spatiale européenne (ESA), l’industrie Thales Alenia Space (TAS) et le laboratoirede l’Intégration du Matériau au Système (IMS) qui développent de nouvelles générations d’unités intégréesde guidage, navigation et pilotage (GNC) avec une fonction de détection des défauts et de tolérance desdéfauts. La mission de référence retenue dans cette thèse est la mission de retour d’échantillons martiens(Mars Sample Return, MSR) de l’ESA. Ce travail se concentre sur la séquence terminale du rendez-vous dela mission MSR qui correspond aux dernières centaines de mètres jusqu’à la capture. Le véhicule chasseurest l’orbiteur MSR (chasseur), alors que la cible passive est un conteneur sphérique. L’objectif au niveaude contrôle est de réaliser la capture avec une précision inférieure à quelques centimètres. Les travaux derecherche traités dans cette thèse s’intéressent au développement des approches sur base de modèle de détectionet d’isolation des défauts (FDI) et de commande tolérante aux défaillances (FTC), qui pourraientaugmenter d’une manière significative l’autonomie opérationnelle et fonctionnelle du chasseur pendant lerendez-vous et, d’une manière plus générale, d’un vaisseau spatial impliqué dans des missions située dansl’espace lointain. Dès lors que la redondance existe dans les capteurs et que les roues de réaction ne sontpas utilisées durant la phase de rendez-vous, le travail présenté dans cette thèse est orienté seulementvers les systèmes de propulsion par tuyères. Les défaillances examinées ont été définies conformément auxexigences de l’ESA et de TAS et suivant leurs expériences. Les approches FDI/FTC présentées s’appuientsur la redondance de capteurs, la redirection de contrôle et sur les méthodes de réallocation de contrôle,ainsi que le FDI hiérarchique, y compris les approches à base de signaux au niveau de capteurs, les approchesà base de modèle de détection/localisation de défauts de propulseur et la surveillance de sécuritéde trajectoire. Utilisant un simulateur industriel de haute-fidélité, les indices de performance et de fiabilitéFDI, qui ont été soigneusement choisis accompagnés des campagnes de simulation de robustesse/sensibilitéMonte Carlo, démontrent la viabilité des approches proposées. / The work addressed in this thesis draws expertise from actions undertaken between the EuropeanSpace Agency (ESA), the industry Thales Alenia Space (TAS) and the IMS laboratory (laboratoirede l’Intégration du Matériau au Système) which develop new generations of integrated Guidance, Navigationand Control (GNC) units with fault detection and tolerance capabilities. The reference mission isthe ESA’s Mars Sample Return (MSR) mission. The presented work focuses on the terminal rendezvoussequence of the MSR mission which corresponds to the last few hundred meters until the capture. Thechaser vehicle is the MSR Orbiter, while the passive target is a diameter spherical container. The objectiveat control level is a capture achievement with an accuracy better than a few centimeter. The research workaddressed in this thesis is concerned by the development of model-based Fault Detection and Isolation(FDI) and Fault Tolerant Control (FTC) approaches that could significantly increase the operational andfunctional autonomy of the chaser during rendezvous, and more generally, of spacecraft involved in deepspace missions. Since redundancy exist in the sensors and since the reaction wheels are not used duringthe rendezvous phase, the work presented in this thesis focuses only on the thruster-based propulsionsystem. The investigated faults have been defined in accordance with ESA and TAS requirements andfollowing their experiences. The presented FDI/FTC approaches relies on hardware redundancy in sensors,control redirection and control re-allocation methods and a hierarchical FDI including signal-basedapproaches at sensor level, model-based approaches for thruster fault detection/isolation and trajectorysafety monitoring. Carefully selected performance and reliability indices together with Monte Carlo simulationcampaigns, using a high-fidelity industrial simulator, demonstrate the viability of the proposedapproaches.

<p>The problem of fault diagnosis consists of detecting and isolating faults present in a system. As technical systems become more and more complex and the demands for safety, reliability and environmental friendliness are rising, fault diagnosis is becoming increasingly important. One example is automotive systems, where fault diagnosis is a necessity for low emissions, high safety, high vehicle uptime, and efficient repair and maintenance.</p><p>One approach to fault diagnosis, providing potentially good performance and in which the need for additional hardware is minimal, is model-based fault diagnosis with residuals. A residual is a signal that is zero when the system under diagnosis is fault-free, and non-zero when particular faults are present in the system. Residuals are typically generated by using a mathematical model of the system and measurements from sensors and actuators. This process is referred to as residual generation.</p><p>The main contributions in this thesis are two novel methods for residual generation. In both methods, systems described by Differential-Algebraic Equation (DAE) models are considered. Such models appear in a large class of technical systems, for example automotive systems. The first method consider observer-based residual generation for linear DAE-models. This method places no restrictions on the model, such as e.g. observability or regularity, in comparison with other previous methods. If the faults of interest can be detected in the system, the output from the design method is a residual generator, in state-space form, that is sensitive to the faults of interest. The method is iterative and relies on constant matrix operations, such as e.g. null-space calculations and equivalence transformations.</p><p>In the second method, non-linear DAE-models are considered. The proposed method belongs to a class of methods, in this thesis referred to as sequential residual generation, which has shown to be successful for real applications. This method enables simultaneous use of integral and derivative causality, and is able to handle equation sets corresponding to algebraic and differential loops in a systematic manner. It relies on a formal framework for computing unknown variables in the model according to a computation sequence, in which the analytical properties of the equations in the model as well as the available tools for equation solving are taken into account. The method is successfully applied to complex models of an automotive diesel engine and a hydraulic braking system.</p>

<p>This thesis investigates the possibility of improving black box fault diagnosis by a process called manifold learning, which simply stated is a way of finding patterns in recorded sensor data. The idea is that there is more information in the data than is exploited when using simple classification algorithms such as k-Nearest Neighbor and Support Vector Machines, and that this additional information can be found by using manifold learning methods. To test the idea of using manifold learning, data from two different fault diagnosis scenarios is used: A Scania truck engine and an electrical system called Adapt. Two linear and one non-linear manifold learning methods are used: Principal Component Analysis and Linear Discriminant Analysis (linear) and Laplacian Eigenmaps (non-linear).Some improvements are achieved given certain conditions on the diagnosis scenarios. The improvements for different methods correspond to the systems in which they are achieved in terms of linearity. The positive results for the relatively linear electrical system are achieved mainly by the linear methods Principal Component Analysis and Linear Discriminant Analysis and the positive results for the non-linear Scania system are achieved by the non-linear method Laplacian Eigenmaps.The results for scenarios without these special conditions are not improved however, and it is uncertain wether the improvements in special condition scenarios are due to gained information or to the nature of the cases themselves.</p>