Bien que le diagnostic des roulements par analyse vibratoire soit un domaine très développé, la recherche sur les représentations parcimonieuses des signaux de vibration est encore nouvelle et difficile pour le diagnostic des machines tournantes. Dans cette thèse, de méthodes nouvelles ont été développées, au moyen de différents modèles stochastiques, associées à des algorithmes efficaces afin de servir l’industrie dans le diagnostic des roulements. Tout d’abord, les modèles parcimonieux présentés dans la littérature sont revus. Les principales publications concernant le diagnostic des machines tournantes ont également été considérées. Enfin, en discutant des avantages et des inconvénients des représentations parcimonieuses, une interprétation des structures creuses d’un point de vue Bayésien est proposée, ce qui donne lieu à deux nouveaux modèles de diagnostic des machines tournantes. Dans un second temps, un nouveau modèle stochastique est proposé : il introduit une variable cachée relative à l’apparition d’impacts et estime le contenu spectral des transitoires correspondants ainsi que le spectre du bruit de fond. Cela donne lieu à un algorithme de détection automatique - sans besoin de pré-filtrage manuel - à partir duquel les fréquences de défaut peuvent être révélées. Le même algorithme permet également de filtrer le signal de défaut de manière très efficace par rapport à d’autres approches basées sur l’hypothèse stationnaire. La performance de l’algorithme est étudiée sur des signaux synthétiques. L’efficacité et la robustesse de la méthode sont également vérifiées sur les signaux de vibration mesurés sur un banc d’essai (engrenages et paliers). Les résultats sont meilleurs ou au moins équivalents à ceux de l’analyse d’enveloppes classique et du kurtogramme rapide. Dans un troisième temps, un nouveau schéma pour l’extraction de signaux cyclostationnaires (CS) est proposé. En considérant la variance périodique en tant que variable cachée, un filtre temporel est conçu de manière à obtenir la reconstruction intégrale des signaux CS caractérisés par une fréquence cyclique préétablie, qui peut être connue à priori ou estimée à partir de la corrélation spectrale. Un intérêt particulier de la méthode est sa robustesse lorsqu’elle est appliquée sur des données expérimentales ainsi qu’une capacité d’extraction supérieure par rapport au filtre de Wiener conventionnel. Finalement, ces exemples expérimentaux témoignent de l’utilisation polyvalente de la méthode à des fins de diagnostic de signaux composés. Pour finir, une analyse comparée utilisant le calcul rapide de la corrélation spectrale est réalisée sur une base de données publiquement disponible et largement utilisée. C’est un point crucial qui fixe un défis non-trivial à résoudre. / Although vibration-based rolling element bearing diagnostics is a very well-developed field, the research on sparse representations of vibration signals is yet new and challenging for machine diagnosis. In this thesis, several novel methods have been developed, by means of different stochastic models, associated with their effective algorithms so as to serve the industry in rolling element bearing diagnostics. First, the sparsity-based model (sparse code, in natural image processing) is investigated based on the current literature. The historical background of sparse representations has been inquired in the field of natural scenes. Along three aspects, its mathematical model with corresponding algorithms has been categorized and presented as a fundamental premise; the main publications are therefore surveyed in the literature on machinery fault diagnosis; finally, an interpretation of sparse structure in the Bayesian viewpoint is proposed which then gives rise to two novel models for machinery fault diagnosis. Second, a new stochastic model is introduced to address this issue: it introduces a hidden variable to indicate the occurrence of the impacts and estimates the spectral content of the corresponding transients together with the spectrum of background noise. This gives rise to an automatic detection algorithm – with no need of manual prefiltering as is the case with the envelope spectrum – from which fault frequencies can be revealed. The same algorithm also makes possible to filter out the fault signal in a very efficient way as compared to other approaches based on the stationary assumption. The performance is investigated on synthetic signals with a high noise-to-signal ratio and also in the case of a mixture of two independent transients. The effectiveness and robustness of the method are also verified on vibration signals measured on a test-bench (gears and bearings). Results are found superior or at least equivalent to those of conventional envelope analysis and fast kurtogram. Third, a novel scheme for extracting cyclostationary (CS) signals is proposed. By regularizing the periodic variance as hidden variables, a time-varying filter is designed so as to achieve the full-band reconstruction of CS signals characterized by some pre-set characteristic frequency. Of particular interest is the robustness on experimental data sets and superior extraction capability over the conventional Wiener filter. It not only deals with the bearing fault at an incipient stage, but it even works for the installation problem and the case of two sources, i.e. bearing and gear faults together. Eventually, these experimental examples evidence its versatile usage on diagnostic analysis of compound signals. Fourth, a benchmark analysis by using the fast computation of the spectral correlation is provided. One crucial point is to move forward the benchmark study of the CWRU data set by uncovering its own unique characteristics.

This thesis deals with thermal management aspects of electric machinery used in high-performance  applications  with  particular  focus put  on electric machines designed for hybrid electric vehicle applications. In the first part of this thesis,  new thermal models of liquid (water and oil) cooled electric machines are proposed.  The proposed thermal models are based on a combination of lumped parameter (LP)  and numerical methods. As  a first  case study,  a permanent-magnet  assisted  synchronous reluctance machine (PMaSRM) equipped with a housing water jacket is considered.  Particular focus is put on the stator winding and a thermal model is proposed that divides the stator slot into a number of elliptical copper and impregna- tion layers.  Additionally, an analysis, using results from a proposed simplified thermal finite element (FE)  model representing only a single slot of the sta- tor and its corresponding end winding, is presented in which the number of layers and the proper connection between the parts of the LP thermal model representing the end winding and the active part of winding are determined. The approach is attractive due to its simplicity  and the fact  that it closely models the actual temperature distribution for common slot geometries.  An oil-cooled induction machine where the oil is in direct contact with the stator laminations  is also considered.  Here, a multi-segment structure is proposed that  divides  the  stator,  winding and cooling  system  into  a number  of an- gular  segments.   Thereby,  the  circumferential  temperature  variation  due to the  nonuniform distribution  of the  coolant  in the  cooling  channels  can be predicted. In the  second part  of this  thesis,  the  thermal  impact  of using  different winding impregnation  and steel  lamination  materials  is  studied.   Conven- tional varnish, epoxy and a silicone based thermally conductive impregnation material are investigated and the resulting temperature distributions in three small induction machines are compared. The thermal impact of using different steel lamination materials is investigated by simulations using the developed thermal  model  of the water  cooled  PMaSRM. The  differences  in alloy con- tents and steel lamination thickness are studied separately and a comparison between the produced iron losses and the resulting hot-spot temperatures is presented. Finally, FE-based approaches  for  estimating  the  induced  magnet  eddycurrent losses in the rotor of the considered PMaSRM are reviewed and compared in the  form  of a case  study  based on simulations.   A  simplified three-dimensional  FE model  and an analytical  model,  both  combined  with time-domain 2D FE analysis, are shown to predict the induced eddy current losses with a relatively good accuracy compared to a complete 3D FE based model.  Hence, the two simplified approaches are promising which motivates a possible future experimental verification. / <p>QC 20130528</p>

S postupom času sa elektrotechnicky průmysl posouval dopředu. Přišel rozvoj v automobilovém průmyslu, leteckém průmyslu, navigácí a dalších odvětvích. S rozvojem automobilového a leteckého průmyslu se zvětšila potřeba pro motory s vyskou spolehlivostí. Tenhle fakt zapříčinil rozvoj motorů s vćero fázami. Tyto stroje mají za úkol zvýšiť spolehlivost ale zároveň zachovat/zvýšit výkon stroje.

LD2668 .T4 1947 R5 / Master of Science

Call number: LD2668 .T4 1960 H35

Call number: LD2668 .T4 1962 C44

Cette dissertation présente un ensemble d'algorithmes visant à en permettre un usage rapide, robuste et automatique des « Support Vector Machines » (SVM) non supervisés dans un contexte d'analyse de données. Les SVM non supervisés se déclinent sous deux types algorithmes prometteurs, le « Support Vector Clustering » (SVC) et le « Support Vector Domain Description » (SVDD), offrant respectivement une solution à deux problèmes importants en analyse de données, soit la recherche de groupements homogènes (« clustering »), ainsi que la reconnaissance d'éléments atypiques (« novelty/abnomaly detection ») à partir d'un ensemble de données. Cette recherche propose des solutions concrètes à trois limitations fondamentales inhérentes à ces deux algorithmes, notamment I) l'absence d'algorithme d'optimisation efficace permettant d'exécuter la phase d'entrainement des SVDD et SVC sur des ensembles de données volumineux dans un délai acceptable, 2) le manque d'efficacité et de robustesse des algorithmes existants de partitionnement des données pour SVC, ainsi que 3) l'absence de stratégies de sélection automatique des hyperparamètres pour SVDD et SVC contrôlant la complexité et la tolérance au bruit des modèles générés. La résolution individuelle des trois limitations mentionnées précédemment constitue les trois axes principaux de cette thèse doctorale, chacun faisant l'objet d'un article scientifique proposant des stratégies et algorithmes permettant un usage rapide, robuste et exempt de paramètres d'entrée des SVDD et SVC sur des ensembles de données arbitraires.

The Norwegian Network Code FIKS from the Norwegian Transmission System Operator (TSO) Statnett, states that synchronous generators ≥ 25 MVA must have a static excitation system. It also includes requirements on the step time response and the available field winding ceiling voltage of the excitation system. An improved brushless excitation system is in operation in some pilot power plants. A rotating thyristor bridge is controlled via Bluetooth. The step time response is as fast as conventional static excitation systems. However, a ceiling voltage factor of 2 requires the thyristor bridge to operate at firing angles about 60 degrees. High torque pulsations, low power factor and low utilization of the exciter is the end result. New power electronic interfaces on the shaft results in a betterutilization of the designed exciter and improves the mechanical performance as well as the controllability of the generator field winding. Permanent magnet rotating exciters increase the field forcing strength of the synchronous generator, yielding improved transient stability (Fault Ride-Through req.). Brushless exciters also reduces regular maintenance of the generator. The thesis includes experiments on a state of the art synchronous generator test setup including constructed PM exciter and different power electronic solutions. Some investigations has been done on industrial power plants as well.

Sequence-specific synthesis is essential to life. In nature, information-rich polymers such as polynucleotides, polypeptides and polysaccharides are responsible for virtually all vital processes. As opposed to nature’s optimised approach towards sequence control employing sophisticated molecular machines such as ribosomes and nucleotide polymerases, the synthetic chemist’s toolbox for the generation of highly ordered monomeric sequences is limited in scope. In this thesis, the realisation of the first artificial small-molecule machines capable of synthesising peptides, translating information that is encoded in a molecular strand, is described. The chemical structure of such machines is based on a rotaxane architecture: a molecular ring threaded onto a molecular axle. The ring carries a reactive arm, a thiolate group that iteratively removes amino acids from the strand that block the path of the macrocycle. The acyl monomers are transferred to a peptide-elongation site through native chemical ligation, thereby translating the information encoded in the track into a growing peptide strand. The synthesis is demonstrated with ~1018 molecular machines acting in parallel; this process generates milligram quantities of a peptide with a single sequence as confirmed by tandem mass spectrometry. Chapter I describes previous strategies that have been employed to realise sequence specific synthesis and gives an overview about relevant literature in the field. Chapter II describes the concept, previous work and model studies which lay the ground work for the more advanced machines. The first generation design of a molecular machine based on transacylation catalysis as well as the second generation design based on native chemical ligation are discussed. The successful operation of a single-barrier rotaxane capable of elongating its reactive arm by a single amide bond formation and subsequent self-immolation is described. Chapter III describes the first small-molecule molecular machine capable of sequence-specific assembly of a tripeptide. The sequence-integrity of the operation product is determined by tandem mass spectrometry and comparison with an authentic sample. Chapter IV describes a novel synthetic approach towards highly complex molecular machines. Using this rotaxane elongation strategy, a molecular machine with four aminoacyl monomers on the strand is reported. The successful operation afforded the expected product resulting from four amide bond forming events without any detectable sequence scrambling.

No description available.