To improve safety, reliability, and efficiency of automotive vehicles and other technical applications, embedded systems commonly use fault diagnosis consisting of fault detection and isolation. Since many systems are constructed as distributed embedded systems including multiple control units, it is necessary to perform global fault isolation using for example a central unit. However, the drawbacks with such a centralized method are the need of a powerful diagnostic unit and the sensitivity against disconnections of this unit. Two alternative methods to centralized fault isolation are presented in this thesis. The first method performs global fault isolation by a istributed sequential computation. For a set of studied systems, themethod gives, compared to a centralizedmethod, amean reduction inmaximumprocessor load on any unitwith 40 and 70%for systems consisting of four and eight units respectively. The second method instead extends the result of the local fault isolation performed in each unit such that the results are globally correct. By only considering the components affecting each specific unit, the extended result in each agent is kept small. For a studied automotive vehicle, the second method gives, compared to a centralized method, a mean reduction in the sizes of the results and the maximum processor load on any unit with 85 and 90% respectively. To perform fault diagnosis, diagnostic tests are commonly used. If the additional evaluation of tests can not improve the fault isolation of a component then the component is ready. Since the evaluation of a test comes with a cost in for example computational resources, it is valuable to minimize the number of tests that have to be evaluated before readiness is achieved for all components. A strategy is presented that decides in which order to evaluate tests such that readiness is achieved with as few evaluations of tests as possible. Besides knowing how fault diagnosis is performed, it is also interesting to assess the effect that fault diagnosis has on for example safety. Since fault tree analysis often is used to evaluate safety, this thesis contributes with a systematic method that includes the effect of fault diagnosis in fault trees. The safety enhancement due to the use of fault diagnosis can thereby be analyzed and quantified.

Fault diagnosis in discrete event systems is studied using a state-based framework. Faults can be either intermittent or permanent. For intermittent faults, system may recover from faulty behaviour through reset. To diagnose such intermittent faults, fault counters are introduced. Fault counters record the number of intermittent faults which must have occurred according to the output observations. This provides the main diagnosis. They also record the number of possible intermittent faults which may have occurred but cannot be confirmed. This provides auxiliary diagnostic information. Fault diagnosability is then studied. Since faults may be intermittent, they may occur repeatedly. Three different notions are studied: 1-diagnosability, 1,k-diagnosability, and 1,infty-diagnosability, and criteria for each of these notions are obtained. The criteria are expressed in terms of fault counters and extend the diagnosability criteria for permanent faults. The concept of a resonant path is introduced, which plays an important role in studying diagnosability.

Fault diagnosis in discrete event systems is studied using a state-based framework. Faults can be either intermittent or permanent. For intermittent faults, system may recover from faulty behaviour through reset. To diagnose such intermittent faults, fault counters are introduced. Fault counters record the number of intermittent faults which must have occurred according to the output observations. This provides the main diagnosis. They also record the number of possible intermittent faults which may have occurred but cannot be confirmed. This provides auxiliary diagnostic information. Fault diagnosability is then studied. Since faults may be intermittent, they may occur repeatedly. Three different notions are studied: 1-diagnosability, 1,k-diagnosability, and 1,infty-diagnosability, and criteria for each of these notions are obtained. The criteria are expressed in terms of fault counters and extend the diagnosability criteria for permanent faults. The concept of a resonant path is introduced, which plays an important role in studying diagnosability.

An interval observer has been illustrated to be a suitable approach to detect and isolate faults affecting complex dynamical industrial systems. Concerning fault detection, interval observation is an appropriate passive robust strategy to generate an adaptive threshold to be used in residual evaluation when model uncertainty is located in parameters (interval model). In such approach, the observer gain is a key parameter since it determines the time evolution of the residual sensitivity to a fault and the minimum detectable fault. This thesis illustrates that the whole fault detection process is ruled by the dynamics of the fault residual sensitivity functions and by the time evolution of the adaptive threshold related to the interval observer. Besides, it must be taken into account that these two observer fault detection properties depend on the used observer gain. As a consequence, the observer gain becomes a tuning parameter which allows enhancing the observer fault detection performance while avoiding some drawbacks related to the analytical models, as the wrapping effect. In this thesis, the effect of the observer gain on fault detection and how this parameter can avoid some observer drawbacks (i.e. wrapping effect) are deeply analyzed. One of the results of this analysis is the determination of the minimum detectable fault function related to a given fault type. This function allows introducing a fault classification according to the fault detectability time evolution: permanently (strongly) detected, non-permanently (weakly) detected or just non-detected. In this fault detection part of this thesis, two examples have been used to illustrate the derived results: a mineral grinding-classification process and an industrial servo actuator. Concerning the interface between fault detection and fault isolation, this thesis shows that both modules can not be considered separately since the fault detection process has an important influence on the fault isolation result. This influence is not only due to the time evolution of the fault signals generated by the fault detection module but also to the fact that the fault residual sensitivity functions determines the faults which are affecting a given fault signal and the dynamics of this fault signal for each fault. This thesis illustrates this point suggesting that the interface between fault detection and fault isolation must consider a set of fault signals properties: binary property, sign property, fault residual sensitivity property, occurrence order property and occurrence time instant property. Moreover, as a result of the influence of the observer gain on the fault detection stage and on the fault residual sensitivity functions, this thesis demonstrates that the observer gain has also a key role in the fault isolation module which might allow enhancing its performance when this parameter is tuned properly (i.e. fault distinguishability may be increased). As a last point, this thesis analyzes the timed discrete-event nature of the fault signals generated by the fault detection module. As a consequence, it suggests using timed discrete-event models to model the fault isolation module. This thesis illustrates that this kind of models allow enhancing the fault isolation result. Moreover, as the monitored system is modelled using an interval observer, this thesis shows as this qualitative fault isolation model can be built up on the grounds of this system analytical model. Finally, the proposed fault isolation method is applied to detect and isolate faults of the Barcelona’s urban sewer system limnimeters. Keywords: Fault Detection, Fault Diagnosis, Robustness, Observers, Intervals, Discrete-event Systems. / En la presente tesis se demuestra que el uso de observadores intervalares para detectar y aislar fallos en sistemas dinámicos complejos constituye una estrategia apropiada. En la etapa de detección del fallo, dicha estrategia permite determinar el umbral adaptativo usado en la evaluación del residuo (robustez pasiva). Dicha metodología, responde a la consideración de modelos con parámetros inciertos (modelos intervalares). En dicho enfoque, la ganancia del observador es un parámetro clave que permite determinar la evolución temporal de la sensibilidad del residuo a un fallo y el mínimo fallo detectable para un tipo de fallo determinado. Esta tesis establece que todo el proceso de detección de fallos viene determinado por la dinámica de las funciones sensibilidad del residuo a los diferentes fallos considerados y por la evolución temporal del umbral adaptativo asociado al observador intervalar. Además, se debe tener en cuenta que estas dos propiedades del observador respecto la detección de fallos dependen de la ganancia del observador. En consecuencia, la ganancia del observador se convierte en el parámetro de diseño que permite mejorar las prestaciones de dicho modelo respecto la detección de fallos mientras que permite evitar algunos defectos asociados al uso de modelos intervalares, como el efecto wrapping. Uno de los resultados obtenidos es la determinación de la función fallo mínimo detectable para un tipo de fallo dado. Esta función permite introducir una clasificación de los fallos en función de la evolución temporal de su detectabilidad: fallos permanentemente detectados, fallos no permanentemente detectados y fallos no detectados. En la primera parte de la tesis centrada en la detección de fallos se utilizan dos ejemplos para ilustrar los resultados obtenidos: un proceso de trituración y separación de minerales y un servoactuador industrial. Respecto a la interfaz entre la etapa de detección de fallos y el proceso de aislamiento, esta tesis muestra que ambos módulos no pueden considerarse separadamente dado que el proceso de detección tiene una importante influencia en el resultado de la etapa de aislamiento. Esta influencia no es debida sólo a la evolución temporal de las señales de fallo generados por el módulo de detección sino también porque las funciones sensibilidad del residuo a los diferentes posibles fallos determinan los fallos que afectan a un determinado señal de fallo y la dinámica de éste para cada uno de los fallos. Esta tesis ilustra este punto sugiriendo que el interfaz entre detección y aislamiento del fallo debe considerar un conjunto de propiedades de dichos señales: propiedad binaria, propiedad del signo, propiedad de la sensibilidad del residuo a un fallo dado, propiedad del orden de aparición de las señales causados por los fallos y la propiedad del tiempo de aparición de estos. Además, como resultado de la influencia de la ganancia del observador en la etapa de detección y en las funciones sensibilidad asociadas a los residuos, esta tesis ilustra que la ganancia del observador tiene también un papel crucial en el módulo de aislamiento, el cual podría permitir mejorar el comportamiento de dicho módulo diseñando éste parámetro del observador de forma adecuada (Ej. Incrementar la distinción de los fallos para su mejor aislamiento). Como último punto, esta tesis analiza la naturaleza temporal de eventos discretos asociada a las señales de fallo generados por el módulo de detección. A consecuencia, se sugiere usar modelos de eventos discretos temporales para modelizar el módulo de aislamiento del fallo. Esta tesis muestra que este tipo de modelos permite mejorar el resultado de aislamiento del fallo. Además, dado que el sistema monitorizado es modelado usando un observador intervalar, esta tesis muestra como este modelo cualitativo de aislamiento puede ser construido usando dicho modelo analítico del sistema. Finalmente, el método propuesto de aislamiento del fallo es aplicado para detectar y aislar fallos en los limnimetros del sistema de alcantarillado de Barcelona. Palabras clave: Detección de Fallos, Diagnosis de Fallos, Robusteza, Observadores, Intervalos, Sistemas de Eventos Discretos.

Tests for detecting faults in analog and mixed-signal circuits have been traditionally derived from the datasheet speci and #64257;cations. Although these speci and #64257;cations describe important aspects of the device, in many cases these application oriented tests are costly to implement and are inefficient in determining product quality. Increasingly, the gap between speci and #64257;cation test requirements and the capabilities of test equipment has been widening. In this work, a systematic method to generate and evaluate alternate tests for detecting parametric faults is proposed. We recognize that certain aspects of analog test generation problem are not amenable to automation. Additionally, functional features of analog circuits are widely varied and cannot be assumed by the test generator. To overcome these problems, an extended device under test (DUT) model is developed that encapsulates the DUT and the DUT speci and #64257;c tasks. The interface of this model provides a well de and #64257;ned and uniform view of a large class of devices. This permits several simpli and #64257;cations in the test generator. The test generator is uses a search-based procedure that requires evaluation of a large number of candidate tests. Test evaluation is expensive because of complex fault models and slow fault simulation techniques. A tester-resident test evaluation technique is developed to address this issue. This method is not limited by simulation complexity nor does it require an explicit fault model. Making use of these two developments, an efficient and automated test generation method is developed. Theoretical development and a number of examples are used to illustrate various concepts that are presented in this thesis.

This thesis presents an on-line particle-filtering-based framework for fault diagnosis and failure prognosis in nonlinear, non-Gaussian systems. The methodology assumes the definition of a set of fault indicators, which are appropriate for monitoring purposes, the availability of real-time process measurements, and the existence of empirical knowledge (or historical data) to characterize both nominal and abnormal operating conditions. The incorporation of particle-filtering (PF) techniques in the proposed scheme not only allows for the implementation of real time algorithms, but also provides a solid theoretical framework to handle the problem of fault detection and isolation (FDI), fault identification, and failure prognosis. Founded on the concept of sequential importance sampling (SIS) and Bayesian theory, PF approximates the conditional state probability distribution by a swarm of points called particles and a set of weights representing discrete probability masses. Particles can be easily generated and recursively updated in real time, given a nonlinear process dynamic model and a measurement model that relates the states of the system with the observed fault indicators. Two autonomous modules have been considered in this research. On one hand, the fault diagnosis module uses a hybrid state-space model of the plant and a particle-filtering algorithm to (1) calculate the probability of any given fault condition in real time, (2) estimate the probability density function (pdf) of the continuous-valued states in the monitored system, and (3) provide information about type I and type II detection errors, as well as other critical statistics. Among the advantages offered by this diagnosis approach is the fact that the pdf state estimate may be used as the initial condition in prognostic modules after a particular fault mode is isolated, hence allowing swift transitions between FDI and prognostic routines. The failure prognosis module, on the other hand, computes (in real time) the pdf of the remaining useful life (RUL) of the faulty subsystem using a particle-filtering-based algorithm. This algorithm consecutively updates the current state estimate for a nonlinear state-space model (with unknown time-varying parameters) and predicts the evolution in time of the fault indicator pdf. The outcome of the prognosis module provides information about the precision and accuracy of long-term predictions, RUL expectations, 95% confidence intervals, and other hypothesis tests for the failure condition under study. Finally, inner and outer correction loops (learning schemes) are used to periodically improve the parameters that characterize the performance of FDI and/or prognosis algorithms. Illustrative theoretical examples and data from a seeded fault test for a UH-60 planetary carrier plate are used to validate all proposed approaches. Contributions of this research include: (1) the establishment of a general methodology for real time FDI and failure prognosis in nonlinear processes with unknown model parameters, (2) the definition of appropriate procedures to generate dependable statistics about fault conditions, and (3) a description of specific ways to utilize information from real time measurements to improve the precision and accuracy of the predictions for the state probability density function (pdf).

This thesis focused on developing a new wavelet for use with the continuous wavelet transform, a new detection method and two de-noising algorithms for rolling element bearing fault signals. The work is based on the continuous wavelet transform and implements a unique Fourier Series estimation algorithm that allows for least squares estimation of arbitrary frequency components of a signal. The final results of the research also included use of the developed detection algorithm for a novel method of estimating the center frequency and bandwidth for use with the industry standard detection algorithm, envelope demodulation, based on actual fault data. Finally, the algorithms and wavelets developed in this paper were tested against seven other wavelet based de-noising algorithms and shown to be superior for the de-noising and detection of inner and outer rolling element race faults.

Electric motors are used widely in industrial manufacturing plants. Bearing faults, insulation faults, and rotor faults are the major causes of electric motor failures. Based on the line current analysis, this dissertation mainly deals with the low cost incipient fault detection of inverter-fed driven motors. Basically, low order inverter harmonics contributions to fault diagnosis, a motor drive embedded condition monitoring method, analysis of motor fault signatures in noisy line current, and a few specific applications of proposed methods are studied in detail. First, the effects of inverter harmonics on motor current fault signatures are analyzed in detail. The introduced fault signatures due to harmonics provide additional information about the motor faults and enhance the reliability of fault decisions. It is theoretically and experimentally shown that the extended fault signatures caused by the inverter harmonics are similar and comparable to those generated by the fundamental harmonic on the line current. In the next chapter, the reference frame theory is proposed as a powerful toolbox to find the exact magnitude and phase quantities of specific fault signatures in real time. The faulty motors are experimentally tested both offline, using data acquisition system, and online, employing the TMS320F2812 DSP to prove the effectiveness of the proposed tool. In addition to reference frame theory, another digital signal processor (DSP)-based phasesensitive motor fault signature detection is presented in the following chapter. This method has a powerful line current noise suppression capability while detecting the fault signatures. It is experimentally shown that the proposed method can determine the normalized magnitude and phase information of the fault signatures even in the presence of significant noise. Finally, a signal processing based fault diagnosis scheme for on-board diagnosis of rotor asymmetry at start-up and idle mode is presented. It is quite challenging to obtain these regular test conditions for long enough time during daily vehicle operations. In addition, automobile vibrations cause a non-uniform air-gap motor operation which directly affects the inductances of electric motor and results quite noisy current spectrum. The proposed method overcomes the challenges like aforementioned ones simply by testing the rotor asymmetry at zero speed.

With the widespread use of doubly-fed induction generators (DFIG) in wind energy conversion systems, condition monitoring is being given importance. Non-intrusive techniques like motor current signature analysis (MCSA), which involves looking for specific frequency components in the current spectrum, are preferred over analysis of magnetic field, temperature, vibrations or acoustic noise which require additional sensors. The major difficulty in MCSA is isolation of the fault, as multiple faults produce similar signatures. Moreover, closed-loop control makes diagnostics more complicated due to inherent compensation by the controller. This thesis presents a method to diagnose static and dynamic air gap eccentricities in doubly-fed induction generators operated for closed-loop stator power control by using a modified control technique to enable detection and isolation of this fault from electrical unbalances in the stator and rotor and load torque oscillations that produce similar effects. The effectiveness of the proposed control is verified using simulations and preliminary experiments performed on a healthy machine.

This dissertation studied new transmission line fault diagnosis approaches using new technologies and proposed a scheme to apply those techniques in preventing and mitigating cascading blackouts. The new fault diagnosis approaches are based on two time-domain techniques: neural network based, and synchronized sampling based. For a neural network based fault diagnosis approach, a specially designed fuzzy Adaptive Resonance Theory (ART) neural network algorithm was used. Several ap- plication issues were solved by coordinating multiple neural networks and improving the feature extraction method. A new boundary protection scheme was designed by using a wavelet transform and fuzzy ART neural network. By extracting the fault gen- erated high frequency signal, the new scheme can solve the difficulty of the traditional method to differentiate the internal faults from the external using one end transmis- sion line data only. The fault diagnosis based on synchronized sampling utilizes the Global Positioning System of satellites to synchronize data samples from the two ends of the transmission line. The effort has been made to extend the fault location scheme to a complete fault detection, classification and location scheme. Without an extra data requirement, the new approach enhances the functions of fault diagnosis and improves the performance. Two fault diagnosis techniques using neural network and synchronized sampling are combined as an integrated real time fault analysis tool to be used as a reference of traditional protective relay. They work with an event analysis tool based on event tree analysis (ETA) in a proposed local relay monitoring tool. An interactive monitoring and control scheme for preventing and mitigating cascading blackouts is proposed. The local relay monitoring tool was coordinated with the system-wide monitoring and control tool to enable a better understanding of the system disturbances. Case studies were presented to demonstrate the proposed scheme. An improved simulation software using MATLAB and EMTP/ATP was devel- oped to study the proposed fault diagnosis techniques. Comprehensive performance studies were implemented and the test results validated the enhanced performance of the proposed approaches over the traditional fault diagnosis performed by the transmission line distance relay.