This thesis develops a methodology for the construction of generalized families of probability distributions in the quantile statistical universe, that is, distributions specified in terms of their quantile functions. The main benefit of the proposed methodology is that it generates quantile-based distributions with skewness-invariant measures of kurtosis. The skewness and kurtosis can therefore be identified and analyzed separately. The key contribution of this thesis is the development of a new type of the generalized lambda distribution (GLD), using the quantile function of the generalized Pareto distribution as the basic building block (in the literature each different type of the GLD is incorrectly referred to as a parameterization of the GLD – in this thesis the term type is used). The parameters of this new type can, contrary to existing types, easily be estimated with method of L-moments estimation, since closed-form expressions are available for the estimators as well as for their asymptotic standard errors. The parameter space and the shape properties of the new type are discussed in detail, including its characterization through L-moments. A simple estimation algorithm is presented and utilization of the new type in terms of data fitting and approximation of probability distributions is illustrated. / Thesis (PhD)--University of Pretoria, 2013. / gm2014 / Statistics / unrestricted

The two sample t-test is the test usually taught in introductory statistics courses to test for the equality of means of two populations. However, the t-test is not the only test available to compare the means of two populations. The randomization test is being incorporated into some introductory courses. There is also the bootstrap test. It is also not uncommon to decide the equality of the means based on confidence intervals for the means of these two populations. Are all those methods equally powerful? Can the idea of non-overlapping t confidence intervals be extended to bootstrap confidence intervals? The powers of seven alternative ways of comparing two population means are analyzed using small samples with data coming from distributions with different degrees of skewness and kurtosis. The analysis is done using simulation; programs in GAUSS were especially written for this purpose.

Ce travail de thèse traite de la détection des défauts mécaniques des roulements à billes par analyse de grandeurs mécaniques et électriques dans des machines synchrones à aimants permanents haute vitesse. Le domaine applicatif de ce travail concerne l'aéronautique. Généralement, pour surveiller l'état des roulements à billes dans un actionneur électrique, des mesures vibratoires sont réalisées. Elles permettent, en exploitant le spectre du signal vibratoire, de mettre facilement en évidence la détérioration du roulement. Cette méthode de surveillance est cependant relativement couteuse en termes d'instrumentation et le placement d'un capteur vibratoire dans des équipements à fort degré d'intégration est parfois difficile. Nous proposons dans ce mémoire d'utiliser d'autres grandeurs physiques prélevées sur le système pour réaliser la surveillance de ces défauts. Il peut s'agir de grandeurs mécaniques (vitesse, position par exemple) et de grandeurs électriques (courant statorique, courant onduleur par exemple). L'utilisation de données déjà disponibles dans l'équipement pour les besoins de la commande permet ainsi de supprimer le système d'acquisition vibratoire. A partir d'enregistrements temporels de données réalisées au cours de campagnes d'essais, nous proposons des méthodologies de traitement du signal permettant d'extraire automatiquement des informations sensibles au défaut à surveiller. L'idée finale est de construire des indicateurs de l'état de santé des roulements permettant de prendre « juste à temps » des décisions fiables relatives à la maintenance ou à la sécurisation de l'équipement. Pour construire ces indicateurs, les signatures spécifiques aux défauts de roulements sont étudiées de manière théorique et expérimentale, pour l'ensemble des grandeurs prélevées. Leurs propriétés sont mises en évidence, permettant ainsi de définir les bandes fréquentielles les plus contributives au diagnostic. L'extraction de ces signatures est réalisée dans le domaine fréquentiel selon plusieurs méthodes. Deux types d'indicateurs automatiques différents sont proposés. Le premier est construit directement à partir du spectre d'amplitude des grandeurs par extraction de l'amplitude des harmoniques dans des bandes fréquentielles particulières. Le second intègre une dimension statistique dans l'analyse en exploitant le caractère aléatoire de certains harmoniques pour détecter la présence du défaut. Des critères de comparaison sont définis et utilisés pour étudier les performances des indicateurs proposés pour deux campagnes d'essais avec des roulements artificiellement dégradés, pour différentes vitesses de fonctionnement et pour différents paramètres de réglage des indicateurs. / This Ph.D. thesis deals with detection of mechanical bearings faults by analysis of mechanical and electrical signals in high speed permanent magnet synchronous machine. The application domain of this work concerns aeronautics. Generally, to monitor the ball bearings status in electrical actuator, the vibration measurements are used. They allow, by extracting the vibration spectrum, to easily detect the deterioration of the bearing. This monitoring method is relatively expensive in terms of instrumentation and placing a vibration sensor in equipment with a high integration degree can be difficult. We propose in this paper to use other physical quantities taken from the system to perform the monitoring of these defects. It may be mechanical quantities (for example speed, position) and electrical quantities (for example stator current, power inverter). From time recording of data carried out during test campaigns, we propose signal processing methodologies to automatically extract information sensitive to the monitored fault. The final idea is to construct indicators of bearings health and make decisions relating to maintenance or equipment security. To construct these indicators, specific bearing defects signatures are studied theoretically and experimentally, for all collected variables. The extraction of these signatures is carried out in frequency domain. Two different types of automatic indicators are proposed. The first is constructed directly from the amplitude spectrum by extraction of the harmonic amplitude of the spectrum in particular frequency bands. The second includes a statistical dimension analysis by exploiting the random nature of some harmonics to detect fault presence. Criteria of comparison are defined and used to study the proposed indicators performances for two trial campaigns with artificially degraded bearings, for different speed functioning and for different regulation of indicators parameters.

Submitted by Renata Lopes (renatasil82@gmail.com) on 2017-04-20T18:23:07Z No. of bitstreams: 1 laryssaramosamado.pdf: 2344885 bytes, checksum: 4328135ddbd0305fc11aa0bf0f8f8b61 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2017-04-24T16:50:29Z (GMT) No. of bitstreams: 1 laryssaramosamado.pdf: 2344885 bytes, checksum: 4328135ddbd0305fc11aa0bf0f8f8b61 (MD5) / Made available in DSpace on 2017-04-24T16:50:29Z (GMT). No. of bitstreams: 1 laryssaramosamado.pdf: 2344885 bytes, checksum: 4328135ddbd0305fc11aa0bf0f8f8b61 (MD5) Previous issue date: 2011-08-29 / CNPq - Conselho Nacional de Desenvolvimento Científico e Tecnológico / A presente dissertação tem como objetivos principais a discussão e a análise do uso de técnicas de monitoramento de espectro aplicadas a sistemas PLC, para que a ocupação deste espectro seja explicitada. Neste contexto, diversas técnicas de processamento de sinais e inteligência computacional são utilizadas para extrair e selecionar o menor número de características que sejam mais representativas para detecção, a fim de projetar o melhor e menos complexo detector de sinais a ser utilizado inicialmente na faixa de frequência entre 1,705 e 100 MHz, mas que permita futuras modificações para aplicações na faixa entre 1,705 e 250 MHz. Além disso, o problema de monitoramento de espectro para sistemas PLC é formalizado, e questões de investigação são analisadas tanto para dados simulados em MATLAB quanto para dados medidos em campo. O processo de medição destes dados é descrito e suas características são explicitadas. Finalmente, a análise dos resultados obtidos indica a adequabilidade das técnicas aplicadas ao problema em questão, porém indicam necessidade do aprofundamento desta investigação. Desta maneira, este trabalho consiste em um estudo inicial sobre importantes questões pertinentes ao monitoramento de espectro de sistemas PLC. / This master thesis aims to discuss and analyze the use of spectrum sensing techniques applied to PLC systems, in order to explicit the spectrum occupation. These techniques extract and select the least quantity of the most representative signal features in order to project the best detector that presents the lowest computational complexity. In addition to that, the spectrum sensing problem is formalized, and a few investigation questions are analyzed for both synthetic and measured data. The measurement of PLC signals and their characterization is also exposed. Although the analysis of the attained results indicate that the techniques used are suitable for the examined problems, their further investigation is necessary, in order to better understand the PLC environment and the spectrum sensing issues related to it. This work is, therefore, an initial study about the mentioned matters.

La rationalité parfaite des investisseurs, base de l'hypothèse d'efficience desmarchés, est de plus en plus discutée. Ceci a conduit au développement de la financecomportementale. Le sentiment de marché, qui en est issu, est l'objet de cette étude.Après l'avoir mis en relation avec la rationalité et défini, ses modes de mesure courantset une évaluation de leur capacité à anticiper les rentabilités sont présentés. Ensuite, autravers de deux recherches largement indépendantes, nous (1) montrons de manièreempirique, essentiellement à partir de modèles multi-Agents et d'une modélisation del'impact des chocs d'information sur la distribution des rentabilités, que les skewness etkurtosis de la distribution des rentabilités peuvent être utilisés comme indicateurs dusentiment de marché ; (2) mettons en évidence la présence de mémoire sur de nombreuxindicateurs de sentiment, ce qui invalide les modalités habituelles de leur utilisation,dans le cadre de stratégies contrarian. / The perfect rationality of investors, one of the foundations of theefficient market hypothesis, is increasingly being questioned. This has led to thedevelopment of behavioral finance. Market sentiment, which stems from it, is the focusof this study. Having first linked this concept to rationality and defined it, this studygoes on to present the most common ways of measuring market sentiment and assesstheir ability to anticipate market returns. Then, using two different studies, we do twothings (1) using mainly multi-Agent models and by modeling the impact of informationshocks on the distribution of returns, we empirically show how skewness and kurtosis inthe distribution of returns can be used as market sentiment indicators; (2) wedemonstrate that many standard sentiment indicators are processes affected by long- orshort-Term memory, making them invalid as contrarian indicators even though this ishow they are typically used.

Prior to 1960's, the fingerprint analysis was carried out manually by human experts and for forensic purposes only. Automated fingerprint identification systems (AFIS) have been developed during the last 50 years. The success of AFIS resulted in that its use expanded beyond forensic applications and became common also in civilian applications. Mobile phones and computers equipped with fingerprint sensing devices for fingerprint-based user identification are common today. Despite the intense development efforts, a major problem in automatic fingerprint identification is to acquire reliable matching features from fingerprint images with poor quality. Images where the fingerprint pattern is heavily degraded usually inhibit the performance of an AFIS system. The performance of AFIS systems is also reduced when matching fingerprints of individuals with large age variations. This doctoral thesis presents contributions within the field of fingerprint image enhancement, segmentation and minutiae detection. The reliability of the extracted fingerprint features is highly dependent on the quality of the obtained fingerprints. Unfortunately, it is not always possible to have access to high quality fingerprints. Therefore, prior to the feature extraction, an enhancement of the quality of fingerprints and a segmentation are performed. The segmentation separates the fingerprint pattern from the background and thus limits possible sources of error due to, for instance, feature outliers. Most enhancement and segmentation techniques are data-driven and therefore based on certain features extracted from the low quality fingerprints at hand. Hence, different types of processing, such as directional filtering, are employed for the enhancement. This thesis contributes by proposing new research both for improving fingerprint matching and for the required pre-processing that improves the extraction of features to be used in fingerprint matching systems. In particular, the majority of enhancement and segmentation methods proposed herein are adaptive to the characteristics of each fingerprint image. Thus, the methods are insensitive towards sensor and fingerprint variability. Furthermore, introduction of the higher order statistics (kurtosis) for fingerprint segmentation is presented. Segmentation of the fingerprint image reduces the computational load by excluding background regions of the fingerprint image from being further processed. Also using a neural network to obtain a more robust minutiae detector with a patch rejection mechanism for speeding up the minutiae detection is presented in this thesis.

Ce travail concerne les techniques d'identification ensembliste dites à erreur inconnue mais bornée. On s'est ici exclusivement intêressé à l'approche ensembliste ellipsoïdale. Dans ce contexte précis, nos contributions portent sur une évaluation réaliste de la borne, basée sur l'analyse du kurtosis du signal d'erreur. On propose également une approche unifiée des algorithmes conduisant à leur formulation numériquement stable, à la différence de la quasi-totalité des solutions publiées. L'analyse de la convergence de ce type de techniques fait intervenir la notion habituelle d'excitation persistante pour laquelle on donne une solution algorithmique permettant de qualifier l'entrée, et comment construire des entrées " optimales ". Enfin on s'est intêressé au compromis que l'on pouvait établir entre algorithme séquentiel sous optimal et une approche globale optimale mais numériquement inaccessible. L'ensemble de ces points est illustré tant en simulation qu'avec des données réelles provenant du monde industriel.

Σκοπός της παρούσας είναι ο έλεγχος της ύπαρξης ημερολογιακών ανωμαλιών στη λειτουργία των διεθνών αγορών, όπως αυτή αποτυπώνεται στην πορεία των χρηματιστηριακών δεικτών κατά την περίοδο 01.01.1988 έως 31.03.2008. Όταν μια ημερολογιακή ανωμαλία γίνεται γνωστή στην αγορά, είναι αναμενόμενο η αντίδραση των επενδυτών στην αναμονή της να βαίνει φθίνουσα με το χρόνο. Για την απόρριψη ή την επιβεβαίωση της υπόθεσης αυτής, στην παρούσα μελέτη, ελέγχεται η ύπαρξη ημερολογιακών ανωμαλιών στις αποδόσεις 15 διεθνών χρηματιστηριακών αγορών τα τελευταία 20 έτη. Κυρίαρχος σκοπός της συγκεκριμένης μελέτης είναι ο έλεγχος της μεταβολής των τάσεων στις προτιμήσεις και τις προσδοκίες των επενδυτών τόσο κατά το πέρασμα των χρόνων όσο και με βάση τη θέση κάθε αγοράς στον παγκόσμιο χάρτη ανάπτυξης. Η εξεταζόμενη περίοδος χωρίζεται σε δύο επιμέρους υποπεριόδους με γνώμονα την διεθνή κρίση που ξεκίνησε στα τέλη του 1999 και επηρέασε καθοριστικά την πορεία της παγκόσμιας οικονομίας. Προσπαθούμε, επομένως, να μελετήσουμε την προσαρμογή των αγορών στις συνθήκες και τα δεδομένα που δημιουργούνται έπειτα από ένα σημαντικό γεγονός. Η μεγάλη βάση του δείγματος - ξεπερνά τα είκοσι έτη, το πλήθος των εξεταζόμενων δεικτών και η προσπάθεια μελέτης διαφορετικών τάσεων με βάση ένα κομβικό σημείο της πορείας της ιστορίας αλλά και με βάση την κατηγοριοποίηση των αγορών ανάλογα με το βαθμό ανάπτυξής τους, διαφοροποιούν την παρούσα μελέτη, και πιστεύουμε ότι αποτελούν ένα ισχυρό κίνητρο για έναν μελετητή ή επενδυτή να αφιερώσει χρόνο στην παρούσα μελέτη. Εξάλλου, τόσο η αξιοσημείωτη διαφοροποίηση των αποτελεσμάτων στις δύο εξεταζόμενες υποπεριόδους, όσο και η χαλαρή τάση που εμφανίζεται σε κάθε κατηγορία αγορών, θεωρούμε ότι δικαιώνουν το εν λόγω εγχείρημα. / When a calendar anomaly becomes acquaintance in the market, the reaction of investors is expected to go declining with the time. For the reject or the confirmation of this affair, in the present study, we check the existence of calendar anomalies in the output of 15 International Stock Exchange markets the last 20 years. The examined period (1988-2008) is separated in two subperiods taking into consideration the international crisis that began in the dues of 1999 in order to be checked the stability of results. The empirical results show important differentiation of results in the two examined subperiods. In any case, the segregation of indicators according to their places in the world market does not appear to attribute substantially conclusions. Specifically in the case of emerging markets, these show that they mark an autonomous movement and to be influenced more by other (internal, mainly) factors.

This thesis concerns the problem of statistical hypothesis testing for mean vector as well as testing for non-normality in a high-dimensional setting which is called the Kolmogorov condition. Since we consider mainly the first and the second moment in testing for mean vector and we utilize the third and the fourth moment in testing for non-normality, this thesis concerns a more general moment testing problem. The research question is related to a data matrix with $p$ rows, which is the number of parameters and $n$ columns which is the sample size, where $p$ can exceed $n$, assuming that the ratio $\frac{p}{n}$ converges when both the number of parameters and the sample size increase.  The first paper reviews the Dempster's non-exact test for mean vector, with a focus on one-sample case. We investigated its size and power properties compared to Hotelling's $\mathit{T}^2$ test as well as Srivastava's test using Monte Carlo simulation.  The second paper concerns the problem of testing for multivariate non-normality in high-dimensional data. We proposed three test statistics which are based on marginal skewness and kurtosis. Simulation studies are carried out for examining the size and power properties of the three test statistics. / Avhandlingen undersöker hypotesprövning i höga dimensioner, under förutsättning att det så kallad Kolmogorovvillkoret (Kolmogorov condition) är uppfyllt. Villkoret innerbär att antalet parametrar ökar tillsammans med storleken på stickprovet med en konstant hastighet. Till kategorin multivariat analys räknas de statistiska metoder som analyserar stickprov från flerdimensionella fördelningar, särskilt multivariat normalfördelning. För högdimensionella data fungerar klassiska skattningar av kovariansmatris inte tillfredställande eftersom komplexiteten med att skatta den inversa kovariansmatrisen ökar när dimensionen ökar. I den första uppsatsen utförs en genomgång av Dempsters (non-exact) test där skattning av den inversa kovariansmatrisen inte behövs. Istället används spåret (trace) av en kovariansmatris. I den andra uppsatsen testas antagandet om normalfördelning med hjälp av tredje och fjärde ordningens moment. Tre olika testvariabler har föreslagits där sumuleringar också presenteras för att jämföra hur väl en icke-normalfördelning identifieras av testet.

Conventional rotating machine vibration monitoring techniques are based on the assumption that changes in the measured structural response are caused by deterioration in the condition of the rotating machine. However, due to changing rotational speed, the measured signal may be non-stationary and difficult to interpret. For this reason, the order tracking technique was introduced. One of main advantages of order tracking over traditional vibration monitoring techniques, lies in its ability to clearly identify non-stationary vibration data, and to a large extent exclude the influences from varying rotational speed. Several order tracking techniques have been developed and researched during the past 20 years. Among these techniques, Fourier Transform Based Order Tracking (FT-OT), Angle Domain Sampling Based Order Tracking (AD-OT) and Vold-Kalman Filter Order Tracking (VKF-OT) are the three most popular techniques and have been commercialised in software. While the VKF-OT is comparatively new, and both its theory and application are different from the other two techniques, the unique advantages of this technique has led to increased research attention in this field. This growing interest in research on the application of the VKF-OT technique on real machines, and its comparative advantages with respect to other order tracking techniques, inspired the present research. With this work, a comprehensive literature of electrical machine condition monitoring was surveyed, which gives a broad perspective of electrical machine monitoring methods ranging through electrical techniques, vibration techniques, temperature techniques and chemical techniques. To simply the process of applying VKF-OT in initial investigations, simulated single-degree-of freedom and two-degree-of freedom rotor models were established, and the application of the VKF-OT technique on these simulated models was explored. Because most of the current research draws significantly on an understanding of the VKF-OT theory, it was also necessary to review and summarize the current status of VKF-OT theory from previous work, as well as explore the procedures for selection of its filter bandwidth when dealing with real data. An experimental set-up for monitoring an electrical alternator was constructed. Real experimental data were subsequently used to compare the advantages and disadvantages of the three popular order tracking techniques. The unique time domain advantage of VKF-OT was implemented, using crest factor and kurtosis values as indictors of the fault condition of the machine. This gave encouraging results. / Dissertation (MSc)--University of Pretoria, 2008. / Mechanical and Aeronautical Engineering / unrestricted