La séparation aveugle de sources à partir de mélanges sous-déterminés se fait traditionnellement en deux étapes: l'estimation des filtres de mélange, puis celle des sources. L'hypothèse de parcimonie temps-fréquence des sources facilite la séparation, qui reste cependant difficile dans le cas de mélanges convolutifs à cause des ambiguités de permutation et de mise à l'échelle. Par ailleurs, la parcimonie temporelle des filtres facilite les techniques d'estimation aveugle de filtres fondées sur des corrélations croisées, qui restent cependant limitées au cas où une seule source est active. Dans cette thèse, on exploite conjointement la parcimonie des sources et des filtres de mélange pour l'estimation aveugle de filtres parcimonieux à partir de mélanges convolutifs stéréophoniques de plusieurs sources. Dans un premier temps, on montre comment la parcimonie des filtres permet de résoudre le problème de permutation, en l'absence de problème de mise à l'échelle. Ensuite, on propose un cadre constitué de deux étapes pour l'estimation, basé sur des versions temps-fréquence de la corrélation croisée et sur la minimisation de norme ℓ1: a) un clustering qui regroupe les points temps-fréquence où une seule source est active; b) la résolution d'un problème d'optimisation convexe pour estimer les filtres. La performance des algorithmes qui en résultent est évalués numériquement sur des problèmes de filtre d'estimation de filtres et de séparation de sources audio.

This dissertation deals with dynamics of engineering structures and principally discusses the identification of the modal parameters (i.e., natural frequencies, damping ratios and vibration modes) using output-only information, the excitation sources being considered as unknown and unmeasurable. To solve these kind of problems, a quite large selection of techniques is available in the scientific literature, each of them possessing its own features, advantages and limitations. One common limitation of most of the methods concerns the post-processing procedures that have proved to be delicate and time consuming in some cases, and usually require good users expertise. The constant concern of this work is thus the simplification of the result interpretation in order to minimize the influence of this ungovernable parameter. A new modal parameter estimation approach is developed in this work. The proposed methodology is based on the so-called Blind Source Separation techniques, that aim at reducing large data set to reveal its essential structure. The theoretical developments demonstrate a one-to-one relationship between the so-called mixing matrix and the vibration modes. Two separation algorithms, namely the Independent Component Analysis and the Second-Order Blind Identification, are considered. Their performances are compared, and, due to intrinsic features, one of them is finally identified as more suitable for modal identification problems. For the purpose of comparison, numerous academic case studies are considered to evaluate the influence of parameters such as damping, noise and nondeterministic excitations. Finally, realistic examples dealing with a large number of active modes, typical impact hammer modal testing and operational testing conditions, are studied to demonstrate the applicability of the proposed methodology for practical applications.

La Diagonalisation Conjointe par Congruence (DCC) d'un ensemble de matrices apparaît dans nombres de problèmes de traitement du signal, tels qu'en Analyse en Composantes Indépendantes (ACI). Les développements récents en ACI sous contrainte de non négativité de la matrice de mélange, nommée ACI semi-non négative, permettent de tirer profit d'une modélisation physique réaliste des phénomènes observés tels qu'en audio, en traitement d'image ou en ingénierie biomédicale. Par conséquent, durant cette thèse, l'objectif principal était non seulement de concevoir et développer des algorithmes d'ACI semi-non négative basés sur de nouvelles méthodes de DCC non négative où la matrice de passage recherchée est non négative, mais également d'illustrer leur intérêt dans le cadre d'applications pratiques de séparation de sources. Les algorithmes de DCC non négative proposés exploitent respectivement deux stratégies fondamentales d'optimisation. La première famille d'algorithmes comprend cinq méthodes semi-algébriques, reposant sur la méthode de Jacobi. Cette famille prend en compte la non négativité par un changement de variable carré, permettant ainsi de se ramener à un problème d'optimisation sans contrainte. L'idée générale de la méthode de Jacobi est de i) factoriser la matrice recherchée comme un produit de matrices élémentaires, chacune n'étant définie que par un seul paramètre, puis ii) d'estimer ces matrices élémentaires l'une après l'autre dans un ordre spécifique. La deuxième famille compte un seul algorithme, qui utilise la méthode des directions alternées. Un tel algorithme est obtenu en minimisant successivement le Lagrangien augmenté par rapport aux variables et aux multiplicateurs. Les résultats expérimentaux sur les matrices simulées montrent un gain en performance des algorithmes proposés par comparaison aux méthodes DCC classiques, qui n'exploitent pas la contrainte de non négativité. Il semble que nos méthodes peuvent fournir une meilleure précision d'estimation en particulier dans des contextes difficiles, par exemple, pour de faibles valeurs de rapport signal sur bruit, pour un petit nombre de matrices à diagonaliser et pour des niveaux élevés de cohérence de la matrice de passage. Nous avons ensuite montré l'intérêt de nos approches pour la résolution de problèmes pratiques de séparation aveugle de sources. Pour n'en citer que quelques-uns, nous sommes intéressés à i) l'analyse de composés chimiques en spectroscopie par résonance magnétique, ii) l'identification des profils spectraux des harmoniques (par exemple, de notes de piano) d'un morceau de musique mono-canal par décomposition du spectrogramme, iii) l'élimination partielle du texte se trouvant au verso d'une feuille de papier fin. Ces applications démontrent la validité et l'intérêt de nos algorithmes en comparaison des méthodes classique de séparation aveugle de source. / The Joint Diagonalization of a set of matrices by Congruence (JDC) appears in a number of signal processing problems, such as in Independent Component Analysis (ICA). Recent developments in ICA under the nonnegativity constraint of the mixing matrix, known as semi-nonnegative ICA, allow us to obtain a more realistic representation of some real-world phenomena, such as audios, images and biomedical signals. Consequently, during this thesis, the main objective was not only to design and develop semi-nonnegative ICA methods based on novel nonnegative JDC algorithms, but also to illustrate their interest in applications involving Blind Source Separation (BSS). The proposed nonnegative JDC algorithms belong to two fundamental strategies of optimization. The first family containing five algorithms is based on the Jacobi-like optimization. The nonnegativity constraint is imposed by means of a square change of variable, leading to an unconstrained problem. The general idea of the Jacobi-like optimization is to factorize the matrix variable as a product of a sequence of elementary matrices which is defined by only one parameter, then to estimate these elementary matrices one by one in a specific order. The second family containing one algorithm is based on the alternating direction method of multipliers. Such an algorithm is derived by successively minimizing the augmented Lagrangian function of the cost function with respect to the variables and the multipliers. Experimental results on simulated matrices show a better performance of the proposed algorithms in comparison with several classical JDC methods, which do not use the nonnegativity as constraint prior. It appears that our methods can achieve a better estimation accuracy particularly in difficult contexts, for example, for a low signal-to-noise ratio, a small number of input matrices and a high coherence level of matrix. Then we show the interest of our approaches in solving real-life problems. To name a few, we are interested in i) the analysis of the chemical compounds in the magnetic resonance spectroscopy, ii) the identification of the harmonically fixed spectral profiles (such as piano notes) of a piece of signal-channel music record by decomposing its spectrogram, iii) the partial removal of the show-through effect of digital images, where the show-through effect were caused by scanning a semi-transparent paper. These applications demonstrate the validity and improvement of our algorithms, comparing with several state-of-the-art BSS methods.

[pt] A medição e classificação de descargas parciais constituem uma importante ferramenta de avaliação dos sistemas de isolamento utilizados em equipamentos de alta tensão. Após o pré-processamento dos dados, que captura, digitaliza e filtra o sinal de descargas parciais, geralmente eliminando os ruídos, existem basicamente duas etapas principais, que são a extração de características e a classificação de padrões. As descargas parciais contêm um conjunto de características discriminatórias únicas que lhes permitem ser reconhecidas. Assim, o primeiro procedimento no processo de classificação é definir quais delas podem ser utilizadas e qual o método de extração destas características. O fenômeno de descargas parciais tem uma natureza transitória e é caracterizado por correntes pulsantes com uma duração de vários nanossegundos até poucos microssegundos. Sua magnitude não é sempre proporcional ao dano causado, sendo que descargas de pequena magnitude podem levar rapidamente à evolução de um defeito. Por isso a necessidade de se entender bem este fenômeno e saber interpretar os dados. Além disso, equipamentos de alta tensão de grande porte, como motores e geradores, podem apresentar mais de uma fonte interna de descargas parciais, sendo importante separar os sinais dessas diferentes fontes antes de realizar a classificação. No caso de outros equipamentos de alta tensão de menor porte, como para-raios e transformadores de corrente de subestação, a simples detecção da presença de descargas parciais interna ao equipamento, independente do número de fontes, já é suficiente para indicar a retirada de operação destes equipamentos, dado seu baixo custo relativo e o elevado grau de importância destes para a confiabilidade do sistema onde estão inseridos. Para um diagnóstico completo e confíável de isolamentos de alta tensão, há a demanda por um sistema de análise capaz de promover com eficácia a detecção de descargas parciais internas aos equipamentos, a separação das diversas fontes de descargas parciais, no caso dos equipamentos de grande porte, bem como realizar a correta classificação do tipo de defeito, com base principalmente na análise das características discriminantes das diferentes fontes e na assinatura dos sinais para os diferentes defeitos. Este estudo contribui para o preenchimento desta lacuna, apresentando metodologias que se mostram robustas e precisas nos testes realizados, de modo que possam efetivamente orientar os especialistas em manutenção na tomada de decisões. Para fazer isso, são propostas novas variáveis capazes de extrair informações relevantes de sinais no tempo medidos em diversos tipos de isolamentos, sendo aplicadas aqui em dados obtidos em campo e em laboratório para avaliar sua eficácia na tarefa. Essas informações são tratadas utilizando técnicas de classificação de padrões e inteligência artificial para determinar de forma automática a presença de descargas parciais, o número de fontes diferentes e o tipo de defeito nos isolamentos de alta tensão utilizados no estudo. Outra contribuição do estudo é a criação de um banco de dados histórico, baseada em processamento de imagem, com padrões de mapas de descargas parciais conhecidos na literatura em máquinas rotativas, para serem utilizados na classificação de novos mapas medidos neste tipo de equipamento. / [en] Measurement and classification of partial discharges are an important tool for the evaluation of insulation systems used in high voltage equipments. After pre-processing of data, which captures, scans and filters the signal of partial discharges, generally eliminating noises, there are basically two main steps, which are the extraction of characteristics and the pattern classification. Partial discharges contain a set of unique discriminatory characteristics that allow them to be recognized. Thus, the first procedure in the classification process is to define which of them can be used and which is the method for extraction of those characteristics. The phenomenon of partial discharges has a transient nature and is characterized by pulsating currents with a duration of several nanoseconds up to a few microseconds. Its magnitude is not always proportional to the damage caused, and discharges of small magnitude can quickly lead to the evolution of a failure. Therefore the need to understand this phenomenon well and to know how to interpret the data. In addition, large high voltage equipments such as motors and generators may have more than one internal source of partial discharges, and it is important to separate the signals from those different sources prior to classification. In the case of smaller high voltage equipments, as surge arrester and substation current transformers, the simple detection of the presence of partial discharges inside the equipment, regardless of the number of sources, is sufficient to indicate the withdrawal of operation of the equipment, given their low relative cost and the high degree of importance of these to the reliability of the system where they are part of. For a complete and reliable diagnosis of high voltage insulations, there is a demand for an analysis system capable of effectively promoting the detection of the partial discharges internal to the equipments, the separation of the various sources of partial discharges in the case of large equipments, as well as to carry out the correct classification of the type of failure. The system should be based mainly on the analysis of the discriminating characteristics of the different sources and the signature of the signals for the different failure. This study contributes to fill this gap by presenting methodologies that are robust and accurate in the tests performed, so that they can effectively guide maintenance specialists in decision making. To do this, new variables are proposed to extract relevant information from time signals measured in various types of insulations, being applied here in field and laboratory data to evaluate their effectiveness in the task. This information is treated using standard classification techniques and artificial intelligence to automatically determine the presence of partial discharges, the number of different sources and the type of defect in the high voltage insulations used in the study. Another contribution of the study is the creation of a historical database, based on image processing, with partial discharge map patterns known in the literature on rotating machines, to be used in the classification of new maps measured in this type of equipment.

La reconstruction de champ source a pour but d’identifier le champ d’excitation en mesurant la réponse du système. Pour l’Holographie acoustique de champ proche (Near-field Acoustic Holography), la réponse du système (pression acoustique rayonnée) est mesurée sur un hologramme bidimensionnel utilisant un réseau de microphones et le champ source (le champ de vitesse acoustique) est reconstruit par une technique de rétropropagation effectuée dans le domaine des nombres d’ondes. L’objectif des travaux présentés est d’utiliser le même type de techniques pour reconstruire le champ de déplacement sur toute la surface d’une plaque en mesurant les vibrations sur des hologrammes à une dimension (lignes de mesures). Dans le domaine vibratoire, l’équation du mouvement de plaque implique la présence de 4 types d’ondes différents, deux étant purement évanescents. Ces derniers peuvent introduire des instabilités dans l’application de la méthode, notamment lorsque les hologrammes sont placés dans le champ lointain des efforts appliqués à la structure. La méthode présentée ici, appelée ”Holographie Vibratoire”, est particulièrement intéressante quand une mesure directe du champ de vitesse est impossible. L’holographie vibratoire permet également de séparer les sources dans le cas d’excitations multiples en les considérant comme des ondes allers ou retours. Il est alors possible d’isoler l’influence de chaque source et de quantifier notamment les champs d’intensités structurales que chacune d’elles génère. L’objectif de cette thèse est de présenter les principes de l’holographie Vibratoire, ses limites, ses applications et de les illustrer par des exemples sur plaque infinie, plaque appuyée et sur des résultats expérimentaux. / The source field reconstruction aims at identifying the excitation field measuring the response of the system. In Near-field Acoustic Holography, the response of the system (the radiated acoustic pressure) is measured on a hologram using a microphones array and the source field (the acoustic velocity field) is reconstructed with a back-propagation technique performed in the wave number domain. The objective of the present works is to use such a technique to reconstruct displacement field on the whole surface of a plate by measuring vibrations on a one-dimensional holograms. This task is much more difficult in the vibratory domain because of the complexity of the equation of motion of the structure. The method presented here and called "Structural Holography" is particularly interesting when a direct measurement of the velocity field is not possible. Moreover, Structural Holography decreases the number of measurements required to reconstruct the displacement field of the entire plate. This method permits to separate the sources in the case of multi-sources excitations by considering them as direct or back waves. It’s possible to compute the structural intensity of one particular source without the contributions of others sources. The aim of this PHD is to present the principles of Structural Holography, its limits, its applications and illustrate them with examples of infinite plate, supported plate and on experimental results.

Dans cette thèse, nous nous sommes attaqués au problème de séparation aveugle de mélanges linéaires de sources ayant des propriétés de cyclo-stationnarité. Trois applications ont été abordées à savoir : télécommunications, vibrations mécaniques et biomécaniques. Dans un premier temps, deux nouvelles méthodes ont été proposées, la première a pour but de séparer aveuglement des sources cyclo-stationnaires partageant une ou plusieurs fréquences cycliques inconnues. Elle combine la diagonalisation conjointe à un nouveau détecteur de points utiles (retard-fréquence cyclique) permettant de composer l’ensemble de matrices de corrélation cyclique devant être diagonalisées conjointement. Quant à la deuxième méthode, elle vise à identifier la matrice de mélange de sources cyclostationnaires de fréquences cycliques inconnues et différentes. L’identification commence par une étape de détection des matrices de rang un, puis décompose en éléments propres le produit de matrices sélectionnées, enfin une méthode de regroupement hiérarchique restitue les colonnes de notre matrice recherchée. Les deux solutions ont été appliquées aux signaux de télécommunications. Dans un second temps, nous avons appliqué d’abord la première méthode proposée sur des signaux mécaniques issus d’un banc de roulements défaillants afin de tester son aptitude à séparer les sources. Ensuite, nous avons proposé une approche qui s’appuie sur l’analyse en composantes parcimonieuses pour séparer les composantes de la force de réaction au sol ayant des propriétés cyclo-stationnaires à l’ordre 1 et 2 / In this thesis, we have tackled the problem of blind separation of linear mixtures of sources with cyclo-stationarity properties. Three applications were studied : telecommunications, mechanical vibrations and biomechanics. First, two new methods have been proposed, the first one aims to blindly separate cyclo-stationary sources sharing one or more unknown cyclic frequencies. It combines the joint diagonalization with a new useful point detector (time lag-cyclic frequency) to compose the set of cyclic correlation matrices to be jointly diagonalized. As for the second method, it aims to identify the mixture matrix of cyclo-stationary sources of unknown and different cyclic frequencies. The identification begins with a step of detecting the matrices of rank one, then the product of selected matrices is decomposed into eigen-elements, and finally a hierarchical regrouping method returns the columns of our sought matrix. Both solutions have been applied to telecommunications signals. In a second step, we first applied the first proposed method on mechanical signals coming from a bank of faulty bearings in order to test its ability to separate the sources. Next, we proposed an approach based on sparse component analysis to separate the components of the ground reaction force with cyclo-stationary properties at order 1 and 2