Le pronostic industriel vise à étendre le cycle de vie d’un dispositif physique, tout en réduisant les couts d’exploitation et de maintenance. Pour cette raison, le pronostic est considéré comme un processus clé avec des capacités de prédiction. En effet, des estimations précises de la durée de vie avant défaillance d’un équipement, Remaining Useful Life (RUL), permettent de mieux définir un plan d’action visant à accroitre la sécurité, réduire les temps d’arrêt, assurer l’achèvement de la mission et l’efficacité de la production.Des études récentes montrent que les approches guidées par les données sont de plus en plus appliquées pour le pronostic de défaillance. Elles peuvent être considérées comme des modèles de type boite noire pour l’ étude du comportement du système directement `a partir des données de surveillance d’ état, pour définir l’ état actuel du système et prédire la progression future de défauts. Cependant, l’approximation du comportement des machines critiques est une tâche difficile qui peut entraîner des mauvais pronostic. Pour la compréhension de la modélisation du pronostic guidé par les données, on considère les points suivants. 1) Comment traiter les données brutes de surveillance pour obtenir des caractéristiques appropriées reflétant l’ évolution de la dégradation? 2) Comment distinguer les états de dégradation et définir des critères de défaillance (qui peuvent varier d’un cas `a un autre)? 3) Comment être sûr que les modèles définis seront assez robustes pour montrer une performance stable avec des entrées incertaines s’ écartant des expériences acquises, et seront suffisamment fiables pour intégrer des données inconnues (c’est `a dire les conditions de fonctionnement, les variations de l’ingénierie, etc.)? 4) Comment réaliser facilement une intégration sous des contraintes et des exigence industrielles? Ces questions sont des problèmes abordés dans cette thèse. Elles ont conduit à développer une nouvelle approche allant au-delà des limites des méthodes classiques de pronostic guidé par les données. / Prognostics and Health Management (PHM) aims at extending the life cycle of engineerin gassets, while reducing exploitation and maintenance costs. For this reason,prognostics is considered as a key process with future capabilities. Indeed, accurateestimates of the Remaining Useful Life (RUL) of an equipment enable defining furtherplan of actions to increase safety, minimize downtime, ensure mission completion andefficient production.Recent advances show that data-driven approaches (mainly based on machine learningmethods) are increasingly applied for fault prognostics. They can be seen as black-boxmodels that learn system behavior directly from Condition Monitoring (CM) data, usethat knowledge to infer its current state and predict future progression of failure. However,approximating the behavior of critical machinery is a challenging task that canresult in poor prognostics. As for understanding, some issues of data-driven prognosticsmodeling are highlighted as follows. 1) How to effectively process raw monitoringdata to obtain suitable features that clearly reflect evolution of degradation? 2) Howto discriminate degradation states and define failure criteria (that can vary from caseto case)? 3) How to be sure that learned-models will be robust enough to show steadyperformance over uncertain inputs that deviate from learned experiences, and to bereliable enough to encounter unknown data (i.e., operating conditions, engineering variations,etc.)? 4) How to achieve ease of application under industrial constraints andrequirements? Such issues constitute the problems addressed in this thesis and have ledto develop a novel approach beyond conventional methods of data-driven prognostics.

Les préoccupations environnementales actuelles nous amènent à envisager des solutions alternatives, telles que la pile à combustible. Cette dernière malgré ses avantages présente des faiblesses qui ralentissent sa diffusion au sein de l'industrie, entre autres, sa trop courte durée de vie. Face à cette considération, nous proposons d'appliquer le PHM à la PEMFC. Il faut donc développer le pronostic puis considérer son insertion au sein d'un système industriel. Nous choisissons de baser l'approche proposée sur un modèle de comportement, tout en proposant de combler le manque de connaissance concernant le vieillissement de la pile par les données, ce qui nous permet amène à développer une approche hybride. Dans ces travaux, le modèle comportemental est étudié sur des durées de plus en plus grandes pour enfin proposer une prédiction de l'ordre du millier d'heure. Afin de prendre en compte une implantation au sein d'un système réel, une étude sur la généricité et applicabilité de l'approche est réalisée. Ainsi, ces travaux proposent une approche de pronostic hybride basée sur un modèle de comportement et étudie son insertion au sein d'un système réel. / The current environmental concerns lead us to consider alternative solutions. The fuel cell can be one of them with numerous advantages, it presents however weaknesses, especially, its life duration which is too short. Face to this issue, we offer to apply the PHM to the PEMFC. For that, it is necessary to develop the prognostics for this application and the possibility of the on-line implementation on an industrial system. It was chosen to base the approach on a behavioral model in which the knowledge gaps are completed with the use of data. So, the approach proposed here, is hybrid. In this work, the behavioral model is studied on laps of time longer in order to finally introduce a prediction of a thousand of hours. Then, the online implementation on a real system is considered with a genericity and an applicability study. This work proposes a hybrid prognostics approach based on a behavioral model and study its implementation on an industrial system.

Spherea (anciennement Cassidian Test & Services), initiateur de la thèse, est un des leaders sur le marché des systèmes automatiques de test (ATE) pour la maintenance des véhicules aéronautiques et de défense. L’intérêt de la société dans la recherche en gestion intégrée de la santé de véhicules est motivé par le taux élevé de fausses déposes d’équipements survenues lors de la maintenance opérationnelle, détectées par les ATE. Ceci engendre des actions de maintenance superflues, et par conséquent des pertes majeurs de temps, d'énergie et d'argent. L’IVHM, de par ses capacités avancées de diagnostic et de pronostic, et son intégration au niveau de l'entreprise de la gestion de santé de véhicules pourrait permettre la réduction des taux de NFF. Néanmoins, les solutions de systèmes IVHM proposées par les communautés scientifique et industrielle sont la plupart du temps développées pour des systèmes spécifiques, et souvent fondées sur des concepts propriétaires. Cela a pour conséquence un manque de consensus à la fois dans les principes structurants des systèmes IVHM et dans leur ingénierie. Aujourd'hui, un défi majeur est de fournir un cadre de modélisation d’IVHM indépendamment du type de système d’intérêt, soutenant l’ingénierie des systèmes IVHM. Vers ce cadre, les principales contributions développées dans cette thèse construisent progressivement les fondations et les piliers d'un cadre de modélisation d’IVHM. La proposition, dans une vision système, des principes structurants d’un système de systèmes permet de définir au général un système IVHM. A partir de cette vision système, le focus de la thèse est orienté sur la gestion de santé du véhicule, fonction de l’IVHM centrée sur le véhicule constituant le catalyseur des décisions de maintenance au niveau opérationnel, ayant la capacité de résoudre le problème industriel à la genèse de la thèse. Les principes structurant clés de cette fonction en trois dimensions (dimension fonctionnelles, dimension d’abstraction, dimension de distribution entre le segment embarqué/sol) sont à la base de la proposition d’un cadre générique de modélisation d’IVHM considérant à la fois les fonctions internes et externes au véhicule. Ce cadre est investigué en cohérence avec une approche construite sur les modèles (MBSE). Conformément à cette approche MBSE, la modélisation, au sein de ce cadre d’IVHM, du module générique de gestion de la santé (gHMM) constitue le support d’intégration des fonctionnalités de diagnostic et de pronostic. Cette modélisation repose sur une vision « boîte noire » et « boîte blanche » du module traduite par un ensemble cohérent de diagrammes SysML, et se basant sur les structures de données standardisées d’OSA-CBM. La formalisation du gHMM permet d'intégrer le diagnostic et le pronostic, processus clés de l’IVHM, dans son sens conventionnel : du diagnostic vers le pronostic, que dans un sens original : du pronostic vers le diagnostic. Ce dernier sens est illustré par la proposition d'un algorithme support à une activité élémentaire du gHMM dans la finalité de réduire les groupes d’ambiguïtés dans le diagnostic. Cette ingénierie aboutit par conséquent à un cadre générique de modélisation, qui par un principe d’instanciation, permet la construction d’une architecture de gestion de l’état de santé d’un système IVHM particulier. Vers cette instanciation la thèse examine les caractéristiques qui impactent la conception d’architectures de gestion de la santé et la sélection d’algorithmes supportant ces architectures, et en propose une formalisation basée sur les ontologies pour la sélection multicritères d’algorithmes de diagnostic et de pronostic appropriés pour la gestion de la santé de véhicules. Finalement, le protocole de validation de l’ensemble des contributions est proposé et illustré à des échelles différentes pour la gestion de l’état de santé d’éoliennes et de drones / Spherea (formerly Cassidian Test & Services), initiator of the PhD thesis, is a leading provider of Automatic Test Equipment (ATE) solutions for aerospace and military vehicles’ maintenance. The company’s interest in Integrated Vehicle Health Management (IVHM) research is motivated by occurrence of No Fault Found (NFF) events detected by ATE, and determining superfluous maintenance activities and consequently major wastes of time, energy and money. IVHM, through its advanced diagnostics and prognostics capabilities, and integration at enterprise level of vehicle health management could solve NFF events occurring during operational-level maintenance. Nevertheless, IVHM systems proposed so far are most of the times developed and matured empirically, for specific vehicle systems, founded on proprietary concepts, and lacking of consensual structuring principles. This results in a lack of consensus in both the structuring principles of IVHM systems and their Systems Engineering. Today, the challenge is to provide an IVHM modelling framework independent from the type of supported system and usable for IVHM Systems Engineering. Towards such framework, the main contributions developed in this thesis progressively build the foundation and pillars of an IVHM modelling framework. The notion of system of systems drives our first proposal of defining principles of an overall IVHM system. From this system vision, the focus of the thesis is oriented on the vehicle centric function of IVHM as catalyst of maintenance decisions at operational level, having the ability to solve the industrial problems at the genesis of the thesis. The key structuring principles of this function are analysed upon three dimensions (functional dimension, a dimension of abstraction, and distribution between the on-board /on-ground segment), setting the basis of the proposal of a generic modelling framework IVHM, considering both vehicle and enterprise centric functions. This framework is built following a Model-based Systems Engineering (MBSE) approach, supported by SysML. The major contribution of the thesis is the modelling, within the framework of IVHM, of the generic Health Management Module (gHMM), support for integration of diagnostics and prognostics, key processes of health management. The gHMM formalization enables to integrate diagnostics and prognostics not only in the conventional way: from diagnosis to prognosis, but also in an original one: from prognostics to diagnostics with the purpose of reducing ambiguity groups; the latter is backed-up through the proposal of an algorithm for one elementary activities of the gHMM. The gHMM MBSE engineering thus leads to a generic modelling framework, which, by a principle of instantiation, allows the construction of an IVHM system designed for the health management of individual vehicle systems. Towards such particularization, the thesis investigates characteristics impacting selection of appropriate supporting algorithms. This analysis enables to identify ten generic macro-criteria, which are further formalized based on ontologies and used within a multi-criteria based methodology suited for selecting diagnostics and prognostics algorithms for vehicle health management. Finally, the validation protocol of the scientific contributions is proposed, and applied at different scales of implementation in the field of wind turbine and UAV health management

As global demand for nuclear energy and threats to nuclear security increase, the need for verification of the peaceful application of nuclear materials and technology also rises. In accordance with the Nuclear Nonproliferation Treaty, the International Atomic Energy Agency is tasked with verification of the declared enrichment activities of member states. Due to the increased cost of inspection and verification of a globally growing nuclear energy industry, remote process monitoring has been proposed as part of a next-generation, information-driven safeguards program. To further enhance this safeguards approach, it is proposed that process monitoring data may be used to not only verify the past but to anticipate the future via prognostic analysis. While prognostic methods exist for health monitoring of physical processes, the literature is absent of methods to predict the outcome of decision-based events, such as the production of undeclared enriched uranium. This dissertation introduces a method to predict the time at which a significant quantity of unaccounted material is expected to be diverted during an enrichment process. This method utilizes a particle filter to model the data and provide a Type III (degradation-based) prognostic estimate of time to diversion of a significant quantity. Measurement noise for the particle filter is estimated using historical data and may be updated with Bayesian estimates from the analyzed data. Dynamic noise estimates are updated based on observed changes in process data. The reliability of the prognostic model for a given range of data is validated via information complexity scores and goodness of fit statistics. The developed prognostic method is tested using data produced from the Oak Ridge Mock Feed and Withdrawal Facility, a 1:100 scale test platform for developing gas centrifuge remote monitoring techniques. Four case studies are considered: no diversion, slow diversion, fast diversion, and intermittent diversion. All intervals of diversion and non-diversion were correctly identified and significant quantity diversion time was accurately estimated. A diversion of 0.8 kg over 85 minutes was detected after 10 minutes and predicted to be 84 minutes and 10 seconds after 46 minutes and 40 seconds with an uncertainty of 2 minutes and 52 seconds.

Optimum Sensor Localization/Selection in A Diagnostic/Prognostic Architecture Guangfan Zhang 107 Pages Directed by Dr. George J. Vachtsevanos This research addresses the problem of sensor localization/selection for fault diagnostic purposes in Prognostics and Health Management (PHM)/Condition-Based Maintenance (CBM) systems. The performance of PHM/CBM systems relies not only on the diagnostic/prognostic algorithms used, but also on the types, location, and number of sensors selected. Most of the research reported in the area of sensor localization/selection for fault diagnosis focuses on qualitative analysis and lacks a uniform figure of merit. Moreover, sensor localization/selection is mainly studied as an open-loop problem without considering the performance feedback from the on-line diagnostic/prognostic system. In this research, a novel approach for sensor localization/selection is proposed in an integrated diagnostic/prognostic architecture to achieve maximum diagnostic performance. First, a fault detectability metric is defined quantitatively. A novel graph-based approach, the Quantified-Directed Model, is called upon to model fault propagation in complex systems and an appropriate figure-of-merit is defined to maximize fault detectability and minimize the required number of sensors while achieving optimum performance. Secondly, the proposed sensor localization/selection strategy is integrated into a diagnostic/prognostic system architecture while exhibiting attributes of flexibility and scalability. Moreover, the performance is validated and verified in the integrated diagnostic/prognostic architecture, and the performance of the integrated diagnostic/prognostic architecture acts as useful feedback for further optimizing the sensors considered. The approach is tested and validated through a five-tank simulation system. This research has led to the following major contributions: ??generalized methodology for sensor localization/selection for fault diagnostic purposes. ??quantitative definition of fault detection ability of a sensor, a novel Quantified-Directed Model (QDG) method for fault propagation modeling purposes, and a generalized figure of merit to maximize fault detectability and minimize the required number of sensors while achieving optimum diagnostic performance at the system level. ??novel, integrated architecture for a diagnostic/prognostic system. ??lidation of the proposed sensor localization/selection approach in the integrated diagnostic/prognostic architecture.

Solid cancers often shed (sub)cellular materials into the circulation, such as circulating tumor cells and extracellular microvesicles. Mounting evidence supports that these circulating materials could serve as surrogate cancer markers for classifying primary tumors, stratifying patients for targeted therapies, assessing treatment efficacy, and achieving clinical benefits. A sensor platform capable of sensitive and portable detection of circulating cancer markers would thus be an invaluable tool, that will advance our understanding of tumor biology as well as clinical outcomes. This dissertation describes various systems that we have developed for quantitative analyses of circulating cancer biomarkers. Firstly, we have developed a novel magnetic resonance sensing platform for microvesicle analyses. By using a chip-based platform that combines microfiltration and bioorthogonal nanoparticle targeting, we demonstrate for the first time that magnetic biosensing can be applied for clinical evaluation of circulating microvesicles in blood samples to monitor cancer therapy. Secondly, we have advanced a new plasmonic sensor to achieve label-free detection of microvesicles. Based on periodic nanohole arrays, this platform has been applied for high-throughput protein profiling of microvesicles in native ascites. Finally, we have implemented microfluidic devices to effectively enrich circulating tumor cells from peripheral whole blood, and to enable comprehensive molecular analyses of isolated tumor cells at a single cell resolution. By enabling rapid, sensitive and cost-effective detection of circulating cancer markers, these developed platforms could significantly expand the reach of preclinical and clinical cancer research, in informing therapy selection, rationally directing trials, and improving sequential monitoring to achieve better clinical outcomes.

Molecular analysis of free circulating (fc)DNA has the potential to change the face of medicine, specifically in cancer diagnostics and in monitoring the efficacy of cancer treatments. In this study, a microfluidic device using AC electrokinetics is developed for rapid concentration and detection of fcDNA from blood. The device concentrates fcDNA using a combination of AC electrothermal flow and dielectrophoresis. The electrothermal fluid motion drives fcDNA towards the center of the electrode where dielectrophoretic trapping occurs. Once fcDNA is collected at the center, the concentration in the sample can be determined by fluorescent analysis using an intercalating dye binding to the double-stranded DNA. Effects of operating parameters are investigated to optimize the device's design. The electrokinetic device isolates high molecular weight DNA and can distinguish from low molecular weight DNA. Quantitative detection of fcDNA in physiologically relevant concentrations is demonstrated toward rapid diagnostics of cancer and monitoring of treatment efficacy.

The ability to forecast machinery failure is vital to reducing maintenance costs, operation downtime and safety hazards. Recent advances in condition monitoring technologies have given rise to a number of prognostic models for forecasting machinery health based on condition data. Although these models have aided the advancement of the discipline, they have made only a limited contribution to developing an effective machinery health prognostic system. The literature review indicates that there is not yet a prognostic model that directly models and fully utilises suspended condition histories (which are very common in practice since organisations rarely allow their assets to run to failure); that effectively integrates population characteristics into prognostics for longer-range prediction in a probabilistic sense; which deduces the non-linear relationship between measured condition data and actual asset health; and which involves minimal assumptions and requirements. This work presents a novel approach to addressing the above-mentioned challenges. The proposed model consists of a feed-forward neural network, the training targets of which are asset survival probabilities estimated using a variation of the Kaplan-Meier estimator and a degradation-based failure probability density estimator. The adapted Kaplan-Meier estimator is able to model the actual survival status of individual failed units and estimate the survival probability of individual suspended units. The degradation-based failure probability density estimator, on the other hand, extracts population characteristics and computes conditional reliability from available condition histories instead of from reliability data. The estimated survival probability and the relevant condition histories are respectively presented as “training target” and “training input” to the neural network. The trained network is capable of estimating the future survival curve of a unit when a series of condition indices are inputted. Although the concept proposed may be applied to the prognosis of various machine components, rolling element bearings were chosen as the research object because rolling element bearing failure is one of the foremost causes of machinery breakdowns. Computer simulated and industry case study data were used to compare the prognostic performance of the proposed model and four control models, namely: two feed-forward neural networks with the same training function and structure as the proposed model, but neglected suspended histories; a time series prediction recurrent neural network; and a traditional Weibull distribution model. The results support the assertion that the proposed model performs better than the other four models and that it produces adaptive prediction outputs with useful representation of survival probabilities. This work presents a compelling concept for non-parametric data-driven prognosis, and for utilising available asset condition information more fully and accurately. It demonstrates that machinery health can indeed be forecasted. The proposed prognostic technique, together with ongoing advances in sensors and data-fusion techniques, and increasingly comprehensive databases of asset condition data, holds the promise for increased asset availability, maintenance cost effectiveness, operational safety and – ultimately – organisation competitiveness.

This thesis project investigates approaches for malfunction prediction using unsupervised, self-organized models, with an orientation on bus fleets. Certain bus malfunctions are not predictable with conventional methods and preventive replacements are too costly and time consuming. Malfunctions that could result in interruption of service or on degradation of safety  are of high priority to predict.The settings of the desired application define the following constraints: definition of a model by an expert is not desirable as it is not a scalable solution, ambient conditions or usage schedule must not affect the prediction, data communication between the systems is limited so data must be compressed with relevant to the problem features. In this work, definition of normal or faulty operation is not handled by an expert, but using the Wisdom of the crowd idea and Consensus Self-organized models for fault detection (COSMO), or by the system's past state by monitoring an autoencoder's reconstruction error. In COSMO each system constructs a model describing its condition and then all distances between models are estimated to find the Most Central Pattern (MCP), which is considered the normal state of the system. The measure of deviation is the tendency of a system's model to be farther from the MCP after a sequence of observations, expressed as a probability that the deviation is incidental.  Factors that apply to the total of systems, such as the weather conditions are thus minimized.The algorithms approach the problem from the scopes of: linear and non linear relations between signals, distribution of values of a single signal, spectrum information of a single signal. This is achieved by constructing relevant models of each observed system (bus). The performance of the implemented algorithms is investigated using ROC curves and real bus fleet data, targeting at predicting a set of malfunctions of the air pressure system.More tests are performed using artificial data with injected malfunctions, to evaluate the performance of the methods. By applying the method on artificial data, the ability of different methods to detect different malfunctions is exhibited.

This dissertation proposes hybrid models for (i) diagnosis and (ii) remaining useful life estimation of a single fatigue crack in a low-pressure turbine blade. The proposed hybrid methods consist of physics-based methods and data-driven methods. In this dissertation, blade tip timing is used to measure the relative tip displacement of a rotor blade. The natural frequency of the blade is determined by detecting the critical speeds of the blade using a newly derived least squares spectral analysis method. The method shares its origin from the Lomb-Scargle periodogram and can detect resonance frequencies in the blade’s displacement while the rotor is in operation. A Campbell diagram is then used to convert the critical speed into a natural frequency. Two kinds of shaft transients are considered, a run-up run-down crossing the same critical speed, is used to test the new method. This dissertation shows that the relative displacement of the blade tip is comparable to those simulated from an analytical single degree of freedom model. It is also shown that the newly proposed resonance detection method estimates the natural frequency of the blade to a high degree of accuracy when compared to the measurements from a modal impact hammer test. The natural frequency obtained from the real time measurement is then used in a pre-constructed hybrid diagnostics model. The diagnostics model provides a probability density function estimation of the surface crack length given the measured natural frequency. A Gaussian Process Regression model is trained on data collected during experiments and finite element simulations of a fatigue crack in the blade. The final part of this dissertation is a sequential inference model for improving the estimation of the crack length and the prediction of the crack growth. The suggested model uses an unscented Kalman filter that improves estimations of the crack length and the rate of crack growth from Paris’ Law coefficients. The model is updated each time a diagnosis is performed on the blade. The RUL of the blade is then determined from an integration of Paris’s Law given the uncertainty estimates of the current damage in the blade. The result of the algorithm is an estimation of the remaining number of cycles to failure. The algorithm is shown to improve the overall estimation of the RUL; however, it is suggested that future work looks at the convergence rate of the method. / Dissertation (MEng)--University of Pretoria, 2019. / Eskom Power Plant Engineering Institute (EPPEI) / Mechanical and Aeronautical Engineering / MEng / Unrestricted