La présente thèse concerne l’étude structurelle pour le recouvrement du défaut par l’approche du bond graph. L'objectif est d'exploiter les propriétés structurelles et causales de l'outil bond graph, afin d’effectuer à la fois le diagnostic et l’analyse de la commande du système physique en présence du défaut. En effet, l’outil bond graph permet de vérifier les conditions structurelles de recouvrement de défauts pas seulement du point de vue de l’analyse de commande, mais aussi en considérant les informations issues de l’étape de diagnostic. Par conséquent, l’ensemble des défauts tolérés est obtenu en mode hors-ligne avant d’effectuer une implémentation réelle. En outre, en estimant le défaut comme une puissance perturbatrice fournie au système, ce qui permet d’étendre les résultats d’analyse structurelle pour le recouvrement du défaut à une compensation locale adaptative, directement à partir du modèle bond graph. Enfin, les résultats obtenus sont validés dans une application d’un véhicule autonome intelligent redondant. / This work deals with structural fault recoverability analysis using the bond graph model. The objective is to exploit the structural and causal properties of the bond graph tool in order to perform both diagnosis and control analysis in the presence of faults. Indeed, the bond graph tool enables to verify the structural conditions of fault recoverability not only from a control perspective but also from a diagnosis one. In this way, the set of faults that can be recovered is obtained previous to industrial implementation. In addition, a novel way to estimate the fault by a disturbing power furnished to the system, enabled to extend the results of structural fault recoverability by performing a local adaptive compensation directly from the bond graph model. Finally, the obtained structural results are validated on a redundant intelligent autonomous vehicle.

La présente thèse concerne l’étude structurelle d’un système modélisé par Bond Graph du point de vue détection et isolation de défaut. En combinant la théorie du module avec les propriétés causales et structurelles de l’outil BG, de simples conditions graphiques permettant de conclure sur la détectabilité et l’isolabilité de défauts composant ont été proposées. Aussi,une procédure de placement de capteur optimal a été présentée pour le recouvrement de la surveillabilité du système en présence de défauts. Pour l’étape de diagnostic, une procédure systématique a été développée pour la génération des indicateurs de défauts sensibles aux défauts composants considérés. L’approche proposée dans ce travail a été validé par une application sur un quart de véhicule autonome intelligent. / This thesis is concerned with system structural analyses for fault detection and isolation. By using the module theory and the causal and structural properties of the Bond Graph tool, simple graphical conditions of structural detectability and isolability of plant faults are proposed. Also, a procedure for optimal sensor placement is performed in order to recover systems monitorability in the presence of plant faults. Moreover, for the diagnosis step, a procedure is proposed for a systematic generation of faults indicator which are only sensitive to the desired plant faults. The proposed approach in this work is validated by an application on a quarter of an autonomous intelligent vehicle.

Le travail de thèse proposé concerne la conception intégrée des algorithmes de surveillance des processus chimiques à base de modèles bond graph. Cette recherche est réalisée dans le cadre de la thématique «Bond graphs pour la supervision des procédés énergétiques» développée entre l’Université de Gabés (Tunisie) et l’Ecole Polytechnique Universitaire de Lille (France). Les processus chimiques sont des procédés polluants et à risque. Ils nécessitent alors pour la protection de l’environnement et du personnel des systèmes de surveillance en ligne pour la détection précoce et l’identification des défaillances. Un processus chimique est le siège de phénomènes de nature diverse : chimiques et ou biochimiques, thermo-fluidique, …. Leur modélisation nécessite alors une approche unifiée. L’outil bond graph par son caractère multidisciplinaire est bien adapté à cette tâche. D’autre part, cet outil peut aussi être utilisé pour la conception d’algorithmes de diagnostic grâce à ses propriétés comportementales, causales et structurelles et déterminer ainsi les conditions de surveillabilité structurelle des équipements pertinents sans calcul numérique. Les principales scientifiques contributions peuvent être résumées comme suit : (i) élaboration d’une bibliothèque de modèles Bond Graph de composants et phénomènes chimiques (ii) méthodologie de diagnostic à base de bond graphs pour la génération d’indicateurs de fautes sensibles à l’apparition des réactions secondaires sources de pollution et d’explosion, (iii) diagnostic robuste aux incertitudes paramétriques du modèle Bond Graphs couplés, (iv) informatisation des procédures d’analyse de surveillabilité d’une classe de procédés chimiques, (v) application à un procédé réel (installation d’estérification). / The proposed Ph.D. thesis deals with integrated design of bond graph model based for health monitoring of chemical processes. This research work is performed within the framework of the topic “Bond graphs for supervision of energetic processes" developed between the University of Gabés (Tunisia) and “Polytech’Lille”, the Engineering School within the University of Science and Technology of Lille. The chemical processes are polluting processes and with risk. They need, for staff safety and environmental protection, online surveillance for early detection and the identification of the failures. chemical processes occur phenomena of various nature: chemical and or biochemical, thermo-fluidic… Their modeling requires a unified approach. The Bond graph as a multidisciplinary tool is well suited to this task.Furthermore, this tool can be used also for the design of diagnosis algorithms thanks to its behavioral, causal and structural properties, and allows providing structural diagnosability conditions of the pertinent equipment without numerical calculation. The main scientific contributions of this research can be summarized as follows: (i)elaboration of a data base of dynamic bond graph models of components and chemical phenomena, (ii) methodology of bond graph model based diagnosis for systematic generation of fault indicators sensitive to the appearance of the secondary reactions source of pollution and explosion, (iii) robust diagnosis based on coupled uncertain bond graph model, (iv) computerization of the diagnosability analysis procedures applied to a class of a chemical process, (v) application to a real process (installation of esterification).

Ce travail de thèse constitue une contribution à la modélisation et au diagnostic des systèmes multi-domaines à commutation (hybrides). Il est appliqué à la supervision des systèmes multi-sources d’énergie propre où l’hydrogène est utilisé comme moyen de stockage. Un tel système associe des composantes énergétiques de nature différente et fait l’objet de commutations produites par la connexion et déconnexion d’un ou plusieurs composants. Ces commutations génèrent différents modes de fonctionnement et sont liées à l’intermittence des sources primaires, aux capacités de stockage et à la disponibilité opérationnelle des ressources matérielles qui constituent le système. La présence de ces commutations engendre une dynamique variable qui est classiquement difficile à exprimer mathématiquement sans exploiter tous les modes. Ces difficultés de modélisation se propagent pour affecter toutes les tâches dépendantes du modèle comme le diagnostic et la gestion de modes de fonctionnement. Pour résoudre ces problématiques, un nouvel outil, Bond Graph Hybride piloté par événements, a été développé. Entièrement graphique, ce formalisme permet une modélisation interdisciplinaire globale du système. En séparant la dynamique continue gérée par le Bond Graph Hybride des états discrets modélisés par un automate intégré, l’approche proposée simplifie la gestion des modes de fonctionnement. Le modèle issu de cette méthodologie est également bien adapté au diagnostic robuste, réalisable sans recourir aux équations analytiques. Associée au diagnostic robuste, cette gestion des modes permet l’implémentation de stratégies de reconfiguration et de protection en présence de défaillances. / This research work constitutes a general contribution towards a simpler modelling and diagnosis of the multidisciplinary hybrid systems. Hybrid renewable energy systems where hydrogen is used to store the surplus of the power fits perfectly under this description. Such system gathers different energetic components that are needed to be connected or disconnected according to different operating conditions. These different switching configurations generate different operating modes and depend on the intermittency of the primary sources, the storage capacities and the operational availability of the different hardwares that constitute the system. The switching behaviour engenders a variable dynamic which is hard to be expressed mathematically without investigating all the operating modes. This modelling difficulty is transmitted to affect all the model-based tasks such as the diagnosis and the operating mode management. To solve this problematic, a new modelling tool, called event-driven hybrid bond graph, is developed. Entirely graphic, this formalism allows a multidisciplinary global modelling for all the operating modes at once. By separating the continuous dynamic driven by the bond graph, from the discrete states handled by an integrated automaton, this approach simplifies the management of the operating modes. The model issued using this methodology is also well-adapted to perform a robust diagnosis which is achievable without referring back to the analytical description of the model. The operating mode management, when associated with the on-line diagnosis, allows the implementation of reconfiguration strategies and protection protocols when faults are detected.

Ce travail de thèse présente différentes contributions pour la génération automatique de modèles représentant les Systèmes Dynamiques Hybrides (SDH) caractérisés par plusieurs modes de fonctionnement. Les composants du système (notamment les capteurs) peuvent être manuellement sélectionnée ou automatiquement exportés à partir des données de Conception Assistée par Ordinateur (CAO) ; ces éléments sont ensuite interconnectés pour reproduire le modèle complet du système industriel. Une fois le modèle créé, des schémas-blocs de simulation et de diagnostic, ainsi que la Matrice de Signature de Fautes (FSM) seront produits. Le logiciel est basé sur les Bonds Graph Hybrides ; la présence de commutations engendre des dynamiques variables (notamment des changements de causalité). Pour lever ces verrous, différents algorithmes sont proposés. En comparaison des logiciels existants, les algorithmes proposés sont valides pour les systèmes continus, discrets ou hybrides. Les théories et algorithmes développés sont appliqués sur un système ferroviaire de freinage électropneumatique. / This thesis work contributes to perform a automed model builder for Hybrid Dynamic Systems (HDS) with numerous modes. Technological components including sensors with an iconic format can be automatically export from a computer-aided design (CAD) scheme or manually drag from database and interconnected, so as to produce the overall HDS model, following industrial technological schemes. Once the model has been created, block diagram for simulation and diagnosis and a Fault Signature Matrix (FSM) could be generated.The theory and algorithm behind the software are based on Hybrid Bond Graphs (HBG). The switching behaviour engenders variables dynamics (particularly causal changes). To solve this problematic, news algorithm are performed. Compared with developed programs for automated modelling, the presented algorithm are valid for continuous, discrete and hybrid systems. The theory is illustrated by an industrial application which consists of the pneumo-electrical control of rolling stock.

Exploring the main factors that determine bond spreads with respect to Treasury rates is one of the most critical issues in the corporate debt market. Credit risk has long been perceived as the most important determinant of bond spreads (Fisher, 1959). One of the most critical parameters in credit risk models is asset volatility, which includes idiosyncratic and systematic components. However, these models do not distinguish between them. Chapter 2 investigates the impact of idiosyncratic volatility on bond portfolio spreads between 2000 and 2010. While the prediction of traditional asset pricing models is that firm-specific risk should be diversified away at aggregate level, I find idiosyncratic volatility plays an incremental role in explaining bond portfolio spreads beyond the market factors. Recovery is an important measurement of credit risk additional to default probability. Chapter 3 focuses on the estimation of firm recovery after bankruptcy using the Leland and Toft (1996) model. Using a large sample of Chapter 11 filings from 1996 to 2007, I find that the recovery derived from the Leland and Toft model has strong explanatory power on the debt recovery observed in the market. Recent literature finds that all extant credit risk models significantly underestimate bond spreads, especially for investment grade bonds of short maturity. Chapter 4 identifies a heretofore ignored component, perceived accounting misstatement, by regressing bond spreads on the proxy of accounting misstatement propensity, while controlling for issuers’ default risk and bond illiquidity risk between January 1994 and June 2002. My thesis deepens the understanding of bond price discovery mechanisms and presents an important challenge for future research to incorporate the strong empirical relationship between idiosyncratic volatility and bond yields in asset pricing models. My thesis also sheds light on the accurate prediction of debt recovery, which is important to the valuation and hedging of risky debt and credit derivatives. Furthermore, my thesis assists in solving the credit spread puzzle by identifying a new risk factor. Overall, my thesis provides new insights into research on the corporate debt market and has important implications for academic scholars and market practitioners. / Thesis (Ph.D, Management) -- Queen's University, 2013-04-30 20:22:12.594

University of Macau / Faculty of Science and Technology / Department of Mathematics

Disulfide bonds play crucial roles to predict the three-dimensional structure and the function of a protein. This thesis develops two algorithms to predict the disulfide bonding state of each cysteine in a protein sequence. These methods are based on the multi-stage framework and the multi-classifier of the support vector machine (SVM). The first algorithm achieves 94.0% accuracy of cysteine state prediction for dataset PDB4136, but in some datasets the results are not as good as our expectation. Thus the second algorithm is designed to improve the predicting ability for the proteins which have oxidized and reduced cysteines simultaneously. In addition, a new training strategy is also developed to increase the prediction accuracy. It appends the probabilities which are obtained from the SVM to the existing features and then starts a new training procedure repeatedly to get better performance. The experiments are performed on the datasets derived from well-known databases, such as Protein Data Bank and SWISS-PROT. It gets 94.3% accuracy for predicting disulfide bonding state on dataset PDB4136, which gets improvement 3.6% compared with the previously best result 90.7%.

The circadian oscillator of the cyanobacterium Synechococcus elongatus is composed of only three proteins, KaiA, KaiB, and KaiC, which together with ATP can generate a self-sustained ~24 hour oscillation of KaiC phosphorylation for several days. KaiA induces KaiC to autophosphorylate whereas KaiB blocks the stimulation of KaiC by KaiA, which allows KaiC to autodephosphorylate. We propose and support a model in which the C-terminal loops of KaiC, the “A-loops”, are the master switch that determines overall KaiC activity. When the A-loops are in their buried state, KaiC is an autophosphatase. When the A-loops are exposed, however, KaiC is an autokinase. The data suggest that KaiA stabilizes the exposed state of the A-loops through direct binding. We also show evidence that if KaiA cannot stabilize the exposed state KaiC remains hypophosphorylated. We propose that KaiB inactivates KaiA by preventing it from stabilizing the exposed state of the A-loops. Thus, KaiA and KaiB likely act by shifting the dynamic equilibrium of the A-loops between exposed and buried states, which shifts the balance of autokinase and autophosphatase activities of KaiC. A-loop exposure likely moves the ATP closer to the sites of phosphorylation and we show evidence in support of how this movement may be accomplished. Density functional theory calculations of isolated Watson–Crick A:U and A:T base pairs predict that adenine 13C2 trans-hydrogen bond deuterium isotope shifts due to isotopic substitution at the pyrimidine H3, 2hΔ13C2, are sensitive to the hydrogen-bond distance between the N1 of adenine and the N3 of uracil or thymine, which supports the notion that 2hΔ13C2 is sensitive to hydrogen-bond strength. Calculated 2hΔ13C2 values at a given N1–N3 distance are the same for isolated A:U and A:T base pairs. Replacing uridine residues in RNA with 5-methyl uridine and substituting deoxythymidines in DNA with deoxyuridines do not statistically shift empirical 2hΔ13C2 values. Thus, we show experimentally and computationally that the C7 methyl group of thymine has no measurable affect on 2hΔ13C2 values. Furthermore, 2hΔ13C2 values of modified and unmodified RNA are more negative than those of modified and unmodified DNA, which supports our hypothesis that RNA hydrogen bonds are stronger than those of DNA. It is also shown here that 2hΔ13C2 is context dependent and that this dependence is similar for RNA and DNA.

The purpose of this study was to investigate possible relationships between school district characteristics and bond underwriter costs for Texas independent school districts. Bond data for all school districts issuing bonds in the five-year period 2004 – 2008 was collected from the Texas Bond Review Board. School district information, including financial, socio- economic/ demographic, debt, and managerial information, was collected from those same districts. The data was analyzed using both descriptive and inferential statistical methods. Descriptive statistics were developed on both bond issue and bond issuer data. Relationships between issue costs and school district characteristics were then examined using multiple regression and factor analysis. Results indicate that, in general, larger districts have an advantage over smaller districts, with underwriter costs generally lower in larger districts. Results also offer modest support for the hypothesis that underwriter fees are related to financial, socio-economic/demographic, debt, and managerial characteristics of school districts.