Les développements récents des techniques IRM engendre des sources de potentiels induits qui « polluent » les signaux électrophysiologiques, utilisé simultanément en IRM pour surveiller le patient et synchroniser les images. Le système élaboré dans ce travail est compose de deux modules « émetteur-récepteur » IRM-compatibles. Le premier permet d’introduire dans le tunnel d’IRM des signaux EPS dont on connait les caractéristiques. Les signaux sont injectés dans un tissu conducteur placé dans le tunnel. Le second module permet de recueillir les signaux après leur contamination par les artéfacts générés par les séquences d’imagerie. Il comporte 20 canaux répartis en quatre bandes fréquentielles (40, 80, 160 et 350 Hz). Des mesures du potentiel induit ont été réalisées en environnements 1,5 T et 3 T. Nous pouvons ainsi analyser les modifications des paramètres des signaux selon les séquences, mais aussi à l’intérieur des différentes bandes de fréquences. Dans ce travail, nous présentons une caractérisation et modélisation des potentiels induits par les commutations de gradients de champ magnétique recueilli par notre dispositif expérimental. La caractérisation et la modélisation permettent d’obtenir des informations pertinentes à prendre en compte pour l’élaboration des algorithmes de filtrage efficaces et robustes. / New developments in MRI techniques create sources of induced voltages that “pollute” the simultaneously acquired electrophysiological signals (EPS), used to monitor patients and/or synchronize images. We developed a device to allow a deep study of the contamination mechanism, which would assist in elaborating new tools to obtain higher quality EPS. The system consists of three main modules: (i) a signal transmission system composed of an EPS generator and a transmission box, which transmits the EPS to a MR-compatible receiver inside the tunnel, (ii) an electro-conductive tissue-mimicking phantom in which the EPS is injected, (iii) a signal collection module composed of a MR compatible amplifier-transmitter that emits, via an optical cable, the collected signal to a receiver box placed outside the MRI room. The receiver box comprises 20 channels distributed into four frequency bands (40, 80, 160, and 350 Hz). Measurements of the induced voltages were performed in 1.5 T and 3 T MRI environments. An algorithm to extract and analyze and model the induced voltages was developed. The modeling algorithm is based on a sinusoidal decomposition of the induced voltages. This work aimed to assess the disturbance level of the EPS, when using larger bandwidth amplifiers. The characterization and modeling of the induced voltages, which represent the dominant noise, reveal relevant information which can be used to develop robust and efficient noise reduction algorithms.

Alternative energy resources (such as photovoltaics, fuel cells, wind turbines, micro-turbines, and internal combustion engines) and energy storage systems (such as batteries, supercapacitors, and flywheels) are increasingly being connected to the utility grid, creating distributed energy resources which require the implementation of an effective distributed power management strategy. Parallel-connected power converters form a critical component in such a distributed energy resources system. This dissertation addresses small-signal modeling and analysis of parallel-connected power converter systems operating in distributed energy environments. This work focuses on DC-DC and DC-AC power converters. First, this work addresses the small-signal modeling and analysis of parallel-connected power converters in a battery/supercapacitor hybrid energy storage system. The small-signal model considers variations in the current of individual energy storage devices and the DC bus voltage as state variables, variations in the power converter duty cycles as control variables, and variations in the battery and the supercapacitor voltages and the load current as external disturbances. This dissertation proposes several different control strategies and studies the effects of variations in controller and filter parameters on system performance. Simulation studies were carried out using the Virtual Test Bed (VTB) platform under various load conditions to verify the proposed control strategies and their effect on the final states of the energy storage devices. Control strategies for single DC-AC three-phase power converters are also identified and investigated. These include a novel PV (active power and voltage) control with frequency droop control loop, PQ (active power and reactive power) control, voltage control, PQ control with frequency droop control, and PQ control with voltage and frequency droop control. Small-signal models of a three-phase power converter system with these control strategies were developed, and the impact of parameter variations on the stability of a PV controlled converter were studied. Moreover, a small-signal model of parallel-connected three-phase DC-AC power converters with individual DC power supplies and network is proposed. The simulations carried out in stand-alone and grid-connected modes verify the combined control strategies that were developed. In addition, a detailed small-signal mathematical model that can represent the zero-sequence current dynamics in parallel-connected three-phase DC-AC power converters that share a single DC power source is presented. The effects of a variety of factors on the zero-sequence current are investigated, and a control strategy to minimize the zero-sequence current is proposed. Time-domain simulation studies verify the results. Simulations of a parallel-connected DC-AC power converter system with nonlinear load were carried out. The active power filter implemented in this system provides sharing of harmonic load between each power converter, and reduces harmonic distortion at the nonlinear load by harmonic compensation.

A aquisição experimental de sinais neuronais é um dos principais avanços da neurociência. Por meio de observações da corrente e do potencial elétricos em uma região cerebral, é possível entender os processos fisiológicos envolvidos na geração do potencial de ação, e produzir modelos matemáticos capazes de simular o comportamento de uma célula neuronal. Uma prática comum nesse tipo de experimento é obter leituras a partir de um arranjo de eletrodos posicionado em um meio compartilhado por diversos neurônios, o que resulta em uma mistura de sinais neuronais em uma mesma série temporal. Este trabalho propõe um modelo linear de tempo discreto para o sinal produzido durante o disparo do neurônio. Os coeficientes desse modelo são calculados utilizando-se amostras reais dos sinais neuronais obtidas in vivo. O processo de modelagem concebido emprega técnicas de identificação de sistemas e processamento de sinais, e é dissociado de considerações sobre o funcionamento biofísico da célula, fornecendo uma alternativa de baixa complexidade para a modelagem do disparo neuronal. Além disso, a representação por meio de sistemas lineares permite idealizar um sistema inverso, cuja função é recuperar o sinal original de cada neurônio ativo em uma mistura extracelular. Nesse contexto, são discutidas algumas soluções baseadas em filtros adaptativos para a simulação do sistema inverso, introduzindo uma nova abordagem para o problema de separação de spikes neuronais. / The experimental acquisition of neuronal signals is a major advance in neuroscience. Through observations of electric current and potential in a brain region, it is possible to understand the physiological processes involved in the action potential generation, and create mathematical models capable of simulating the behavior of the neuronal cell. A common practice in this kind of experiment is to obtain readings from an array of electrodes positioned in a medium shared by several neurons, which results in a mixture of neuronal signals in the same time series. This work proposes a discrete-time linear model of the neuronal signal during the firing of the cell. The coefficients of this model are estimated using real samples of the neuronal signals obtained in vivo. The conceived modeling process employs system identification and signal processing concepts, and is dissociated from any considerations about the biophysical function of the neuronal cell, providing a low-complexity alternative to model the neuronal spike. In addition, the use of a linear representation allows the idealization of an inverse system, whose main purpose is to recover the original signal of each active neuron in a given extracellular mixture. In this context, some solutions based on adaptive filters are discussed for the inverse model simulation, introducing a new approach to the problem of neuronal spike separation.

<p dir="ltr">The industry's interest in electrified powertrain-equipped vehicles has increased due to environmental and economic reasons. Electrified powertrains, in general, produce lower sound and vibration level than those equipped with internal combustion engines, making noise and vibration (N&V) from other non-engine powertrain components more perceptible. One such N&V type that arouses concern to both vehicle manufacturers and passengers is gear growl, but the signal characteristics of gear growl noise and vibration and the threshold of those characteristics that can be used to determine whether a gear growl requires attention are not yet well understood. This study focuses on developing a method to detect gear-growl based on the N\&V measurements and determining thresholds on various severities of gear-growl using supervised machine learning classification. In general, a machine learning classifier requires sufficient high-quality training data with strong information independence to ensure accurate classification performance. In industrial practices, acquiring high-quality vehicle NVH data is expensive in terms of finance, time, and effort. A physically informed data augmentation method is, thus, proposed to generate realistic powertrain NVH signals based on high-quality measurements which not only provides a larger training data set but also enriches the signal feature variations included in the data set. More specifically, this method extracts physical information such as angular speed, tonal amplitudes distribution, and broadband spectrum shape from the measurement data. Then, it recreates a synthetic signal that mimics the measurement data. The measured and simulated (via data augmentation) are transformed into feature matrix representation so that the N\&V signals can be used in the classification model training process. Features describing signal characteristics are studied, extracted, and selected. While the root-mean-square (RMS) of the vibration signal and spectral entropy were sufficient for detecting gear-growl with a test accuracy of 0.9828, the acoustic signal required more features due to background noise, making data linearly inseparable. The minimum Redundancy Maximum Relevance (mRMR) feature scoring method was used to assess the importance of acoustic signal features in classification. The five most important features based on the importance score were the angular acceleration of the driveshaft, the time derivative of RMS, the tone-to-noise ratio (TNR), the time derivative of the spectral spread of the tonal component of the acoustic signal, and the time derivative of the spectral spread of the original acoustic signal (before tonal and broadband separation). A supervised classification model is developed using a support vector machine from the extracted acoustic signal features. Data used in training and testing consists of steady-state vehicle operations of 25, 35, 45, and 55 mph, with two vehicles with two different powertrain specs: axles with 4.56 and 6.14 gear ratios. The dataset includes powertrains with swapped axles (four different configurations). Techniques such as cost weighting, median filter, and hyperparameter tuning are implemented to improve the classification performance where the model classifies if a segment in the signal represents a gear-growl event or no gear-growl event. The average accuracy of test data was 0.918. A multi-class classification model is further implemented to classify different severities based on preliminary subjective listening studies. Data augmentation using signal simulation showed improvement in binary classification applications. In this study, only gear-growl was used as a fault type. Still, data augmentation, feature extraction and selection, and classification methods can be generalized for NVH signal-based fault diagnosis applications. Further listening studies are suggested for improved classification of multi-class classification applications.</p>

No description available.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Piezoelectric transformers (PTs) allow the design of promising power supply applications, increasing efficiency, reducing size, facilitating the achievement of high transformation ratio, besides providing high immunity against electromagnetic noise. Due to the electrical equivalent model having resonant characteristics, some resonant topologies are naturally suitable for these power supplies, i.e. the Class- E, Half-Bridge, Full-Bridge and Push-pull. Among these topologies, the Class-E converter has a highlight of having one controlled switch. The static gain of the Class-E is changed through the switching frequency variation, while the duty cycle is adjusted with the purpose of achieving soft switching for different switching frequencies and loads. The analisys of this process becomes complex when the system has a high number of reactive elements. One way to simplify this analisys is applying a normalized methodology. On this regard, the first result of this work is the normalized analisys of the functionally of the Class-E converter, including normalized load and switching frequency variation. This allows choosing one optimum point for the static design, without the necessity of design parameters. The main objective of this analisys is the obtention of the duty cycle behavior in order to have soft switching for all operation points. In a second moment, a small-signal model was derived using the generalized averaging method, through Fourier series aproximation. The model describes the relevant poles and zeros of the system, being accurate enough for different loads and switching frequencies. The behavior of resonant converters changes considerably for different operating points; therefore it is important to have a model that represents the system well. The normalized analisys allowed simplifying the small-signal model derivation, once soft switching is achieved for all the operation points. Experimental measurements validate either the normalized or the small signal derivation methodologies. The measurements were achieved for a 3W step-down converter, with universal 85-265 V AC input and 6 V DC output. / Os Transformadores Piezoelétricos (PTs) permitem o projeto de aplicações promissoras para fontes de alimentação até 100W, melhorando a eficiência, reduzindo o tamanho, facilitando a obtenção de grandes relações de transformação, além de proporcionar alta imunidade contra ruídos eletromagnéticos e interferências. Os PTs apresentam modelo elétrico ressonante, trazendo a necessidade de implementação juntamente com topologias de conversores ressonantes, como por exemplo os conversores: Classe-E, Meia Ponte, Ponte Completa e Push-pull. Dentre estas topologias, o conversor Classe-E se destaca por apresentar somente um interruptor controlado. O ganho estático do conversor Classe-E é obtido através da variação da freqüência de chaveamento, e a razão cíclica muda para atender as condições de comutação suave para diferentes freqüências e cargas. A análise deste processo se torna complexa à medida que o sistema apresenta inúmeros elementos reativos. Uma maneira de simplificar esta análise é utilizar uma metodologia normalizada. Devido a isso, o primeiro resultado deste trabalho é a análise normalizada do funcionamento do conversor piezoelétrico Classe- E, incluindo variação normalizada da frequênciade operação e da carga. Isso permite escolher um ponto ótimo de projeto estático, sem a necessidade de parâmetros de projeto. O objetivo principal desta análise normalizada é a obtenção do comportamento da razão cíclica para obter comutação suave em todos os pontos de operação. Em um segundo momento, um modelo de pequenos sinais foi derivado utilizando a metodologia do modelo médio generalizado, através de aproximação por series de Fourier. O modelo descreve os pólos e zeros relevantes do sistema, sendo suficientemente preciso para diferentes cargas e da frequencias de operação. O comportamento de conversores ressonantes varia consideravelmente para diferentes pontos de operação, pois isso um modelo que permita avaliar estes pontos de maneira precisa se faz importante. A análise normalizada permitiu simplificar a derivação do modelo de pequenos sinais, uma vez que garante a operação em comutação suave. Para validar a metodologia apresentada, são mostrados resultados experimentais para um conversor abaixador de 3W, entrada universal de 85-260 V AC e saída de 6 V DC.

De nombreux traitements appliqués aux signaux audio travaillent sur une représentation Temps-Fréquence (TF) des données. Lorsque le résultat de ces algorithmes est un champ spectral d’amplitude, la question se pose, pour reconstituer un signal temporel, d’estimer le champ de phase correspondant. C’est par exemple le cas dans les applications de séparation de sources, qui estiment les spectrogrammes des sources individuelles à partir du mélange ; la méthode dite de filtrage de Wiener, largement utilisée en pratique, fournit des résultats satisfaisants mais est mise en défaut lorsque les sources se recouvrent dans le plan TF. Cette thèse aborde le problème de la reconstruction de phase de signaux dans le domaine TF appliquée à la séparation de sources audio. Une étude préliminaire révèle la nécessité de mettre au point de nouvelles techniques de reconstruction de phase pour améliorer la qualité de la séparation de sources. Nous proposons de baser celles-ci sur des modèles de signaux. Notre approche consiste à exploiter des informations issues de modèles sous-jacents aux données comme les mélanges de sinusoïdes. La prise en compte de ces informations permet de préserver certaines propriétés intéressantes, comme la continuité temporelle ou la précision des attaques. Nous intégrons ces contraintes dans des modèles de mélanges pour la séparation de sources, où la phase du mélange est exploitée. Les amplitudes des sources pourront être supposées connues, ou bien estimées conjointement dans un modèle inspiré de la factorisation en matrices non-négatives complexe. Enfin, un modèle probabiliste de sources à phase non-uniforme est mis au point. Il permet d’exploiter les à priori provenant de la modélisation de signaux et de tenir compte d’une incertitude sur ceux-ci. Ces méthodes sont testées sur de nombreuses bases de données de signaux de musique réalistes. Leurs performances, en termes de qualité des signaux estimés et de temps de calcul, sont supérieures à celles des méthodes traditionnelles. En particulier, nous observons une diminution des interférences entre sources estimées, et une réduction des artéfacts dans les basses fréquences, ce qui confirme l’intérêt des modèles de signaux pour la reconstruction de phase. / A variety of audio signal processing techniques act on a Time-Frequency (TF) representation of the data. When the result of those algorithms is a magnitude spectrum, it is necessary to reconstruct the corresponding phase field in order to resynthesize time-domain signals. For instance, in the source separation framework the spectrograms of the individual sources are estimated from the mixture ; the widely used Wiener filtering technique then provides satisfactory results, but its performance decreases when the sources overlap in the TF domain. This thesis addresses the problem of phase reconstruction in the TF domain for audio source separation. From a preliminary study we highlight the need for novel phase recovery methods. We therefore introduce new phase reconstruction techniques that are based on music signal modeling : our approach consists inexploiting phase information that originates from signal models such as mixtures of sinusoids. Taking those constraints into account enables us to preserve desirable properties such as temporal continuity or transient precision. We integrate these into several mixture models where the mixture phase is exploited ; the magnitudes of the sources are either assumed to be known, or jointly estimated in a complex nonnegative matrix factorization framework. Finally we design a phase-dependent probabilistic mixture model that accounts for model-based phase priors. Those methods are tested on a variety of realistic music signals. They compare favorably or outperform traditional source separation techniques in terms of signal reconstruction quality and computational cost. In particular, we observe a decrease in interferences between the estimated sources and a reduction of artifacts in the low-frequency components, which confirms the benefit of signal model-based phase reconstruction methods.