The Self-Organizing Map (SOM) is a type of neural network that forms a regular grid of neurons where clusters of neurons represent different classes of targets. The aim of this thesis is to design electromagnetic target classifiers by using the Self-Organizing Map (SOM) type artificial neural networks for dielectric and conducting objects with simple or complex geometries. Design simulations will be realized for perfect dielectric spheres and also for small-scaled aircraft targets modeled by thin conducting wires. The SOM classifiers will be designed by target features extracted from the scattered signals of targets at various aspects by using the Wigner distribution. Noise performance of classifiers will be improved by using slightly noisy input data in SOM training.

This thesis applies an electromagnetic feature extraction technique to design electromagnetic target classifiers for conductors, dielectrics and dielectric coated conductors using their natural resonance related late-time scattered responses. Classifier databases contain scattered data at only a few aspects for each candidate target. The targets are dielectric spheres of varying sizes and refractive indices, perfectly conducting spheres varying sizes and dielectric coated conducting spheres of varying refractive indices and thickness in coating. The applied classifier design technique is suitable for real-time target classification because of the computational efficiency of feature extraction and decision making approaches. The Wigner-Ville Distribution (WD) is employed in this study in addition to the Principal Components Analysis (PCA) technique to extract target features mainly from late-time target responses. WD is applied to the back-scattered responses at different aspects. To decrease aspect dependency, feature vectors are extracted from selected late-time portions of the WD outputs that include natural resonance related information. Principal components analysis is also used to fuse the feature vectors and/or late-time target responses extracted from reference aspects of a given target into a single characteristic feature vector for each target to further reduce aspect dependency.

Les entraînements électriques à base de machine asynchrone sont largement utilisés dans les applications industrielles en raison de leur faible coût, de leurs performances et de leur robustesse. Cependant, des modes de fonctionnement dégradés peuvent apparaître durant la vie de la machine. L'une des raisons principales de ces défaillances reste les défauts de roulements à billes. Afin d'améliorer la sûreté de fonctionnement des entraînements, des schémas de surveillance peuvent être mis en place afin d'assurer une maintenance préventive. Ce travail de thèse traite de la détection et du diagnostic des défauts mécaniques et plus particulièrement des défauts de roulements dans une machine asynchrone. Généralement, une surveillance vibratoire peut être mise en place. Cette méthode de surveillance est cependant souvent chère du fait de la chaîne de mesure. Une approche, basée sur l'analyse et le traitement des courants statoriques, est alors proposée, afin de suppléer à l'analyse vibratoire. L'étude est basée sur l'existence et la caractérisation des effets des oscillations du couple de charge sur les courants d'alimentation. Un schéma de détection est alors introduit pour détecter différents types de défauts de roulements. De plus, des variables mécaniques, telles que la vitesse ou le couple, sont également reconstruites afin de fournir une indication sur la présence de défauts de roulements. Par ailleurs, un diagnostic des modulations des courants statoriques est proposé, en régime permanent et en régime transitoire, quel que soit le rapport entre les fréquences porteuse et modulante. Les méthodes étudiées sont la transformée de Hilbert, la transformée de Concordia, l'amplitude et la fréquence instantanées ainsi que la distribution de Wigner-Ville.

Most signal processing methods were developed for continuous signals. Digital devices, such as the computer, process only discrete signals. This dissertation proposes new techniques to accurately define and efficiently implement an important signal processing method---the time--frequency distribution (TFD)---using discrete signals. The TFD represents a signal in the joint time--frequency domain. Because these distributions are a function of both time and frequency they, unlike traditional signal processing methods, can display frequency content that changes over time. TFDs have been used successfully in many signal processing applications as almost all real-world signals have time-varying frequency content. Although TFDs are well defined for continuous signals, defining and computing a TFD for discrete signals is problematic. This work overcomes these problems by making contributions to the definition, computation, and application of discrete TFDs. The first contribution is a new discrete definition of TFDs. A discrete TFD (DTFD) should be free from the sampling-related distortion known as aliasing and satisfy all the important mathematical properties that the continuous TFD satisfies. Many different DTFD definitions exist but none come close to attaining this ideal. I propose three new components which make up the DTFD: 1) a new discrete Wigner--Ville distribution (DWVD) definition which satisfies all properties, 2) a new discrete analytic signal which minimises aliasing in the DWVD, and 3) a new method to define and convolve the discrete kernel with the DWVD to produce the DTFD. The result: a DTFD definition that, relative to the existing definitions, better approximates the ideal DTFD. The second contribution is two sets of computationally efficient algorithms to compute the proposed DTFD. The first set of algorithms computes the DTFD exactly; the second set requires less memory than the first set by computing time- and, or frequency-decimated versions of the DTFD. Both sets of algorithms reduce the computational load by exploiting symmetries in the DTFD and by constructing kernel-specific algorithms for four different kernel types. The third, and final, contribution is a biomedical application for the proposed DTFD and algorithms. This application is to accurately detect seizure events in newborn electroencephalogram (EEG) signals. Existing detection methods do not perform well enough for use in a clinical setting. I propose a new method which is more robust than existing methods and show how using the proposed DTFD, comparative to an existing DTFD, improves detection performance for this method. In summary, this dissertation makes practical contributions to the area of time--frequency signal processing by proposing an improved DTFD definition, efficient DTFD algorithms, and an improved newborn EEG seizure detection method using DTFDs.

Ce travail de thèse traite de la détection et du diagnostic de défaillances mécaniques par analyse du courant statorique dans les entraînements électriques à base de machine asynchrone. Deux effets d'un défaut mécanique, des oscillations de couple et une excentricité d'entrefer, sont supposés. La modélisation par approche des ondes de forces magnétomotrices et de perméance conduit à deux modèles analytiques du signal courant. La conséquence des défauts est soit une modulation de phase, soit une modulation d'amplitude du signal courant statorique. Ces phénomènes sont détectés par une analyse spectrale en régime permanent, ou des méthodes temps fréquence en régime transitoire. Les méthodes étudiées sont la fréquence instantanée, le spectrogramme et la représentation de Wigner-Ville. L'estimation paramétrique d'indices de modulation a également été traitée. Des résultats de simulation et expérimentaux permettent de valider les signatures et d'extraire de façon automatique des indicateurs de défaut. De plus, une méthode permettant la distinction des oscillations de couple d'une excentricité dynamique est proposée. L'étude est complétée par une implémentation sur DSP des méthodes temps-fréquence afin de démontrer la faisabilité d'une surveillance en ligne.

L'indexation de documents numérisés manuscrits pose le problème de la segmentation en lignes qui, si elle échoue, handicape les étapes suivantes d'extraction et de reconnaissance de mots. Dans les documents arabe anciens, s'ajoute à ce problème, la présence dans les marges, d'annotations souvent composées de lignes obliques. La détection de ces lignes est nécessaire et constitue un défi important pour l'indexation de ces documents. Ainsi, la segmentation visée dans ce travail de thèse concerne l'extraction de lignes multi-orientées. Pour ce problème, la bibliographie ne présente que des techniques rudimentaires basées essentiellement sur une projection directe de l'image du document suivant une seule direction et donc non applicable à du texte multi-orienté. Devant ce manque, nous avons proposé une approche adaptative permettant de localiser d'abord les zones d'orientation différentes, puis de s'appuyer sur chaque orientation locale pour extraire les lignes. Pendant ma thèse, j'ai développé les points suivants : – Application d'un maillage automatique en utilisant le modèle de contour actif (snake). – Préparation du signal de profil de projection en supprimant tous les pixels qui ne sont pas nécessaires dans le calcul de l'orientation. Ensuite, application de toutes les distributions d'énergie de la classe de Cohen sur le profil de projection pour trouver la meilleure distribution qui donne l'orientation. – Application de quelques règles d'extension pour trouver les zones. – Extraction des lignes en se basant sur un algorithme de suivi des composantes connexes. – Séparation de lignes

The development of low-frequency SONAR systems, using a network of autonomous systems in unmanned vehicles, provides a practical means for bistatic measurements (i.e. when the source and receiver are widely separated, thus allowing multiple viewpoints of a target). Furthermore, time-frequency analysis, in particular Wigner-Ville analysis, takes advantage of the evolution of the time dependent echo spectrum to differentiate a man-made target (e.g. an elastic spherical shell, or cylinder) from a natural one of the similar shape (e.g. a rock). Indeed, key energetic features of man-made objects can aid in identification and classification in the presence of clutter and noise. For example, in a fluid-loaded thin spherical shell, an energetic feature is the mid-frequency enhancement echoes (MFE) that result from antisymmetric Lamb waves propagating around the circumference of the shell, which have been shown to be an acoustic feature useful in this pursuit. This research investigates the enhancement and benefits of bistatic measurements using the Wigner-Ville analysis along with acoustic imaging methods. Additionally, the advantage of joint space-time-frequency coherent processing is investigated for optimal array processing to enhance the detection of non-stationary signals across an array. The proposed methodology is tested using both numerical simulations and experimental data for spherical shells and solid cylinders. This research was conducted as part of the Shallow Water Autonomous Mine Sensing Initiative (SWAMSI) sponsored by ONR.

Brushless Direct Current (BLDC) motors are one of the motor types rapidly gaining popularity. BLDC motors are being increasingly used in critical high performance industries such as appliances, automotive, aerospace, consumer, medical, industrial automation equipment and instrumentation. Fault detection and condition monitoring of BLDC machines is therefore assuming a new importance. The objective of this research is to advance the field of rotor and load fault diagnosis in BLDC machines operating in a variety of operating conditions ranging from constant speed to continuous transient operation. This objective is addressed as three parts in this research. The first part experimentally characterizes the effects of rotor faults in the stator current and voltage of the BLDC motor. This helps in better understanding the behavior of rotor defects in BLDC motors. The second part develops methods to detect faults in loads coupled to BLDC motors by monitoring the stator current. As most BLDC applications involve non-stationary operating conditions, the diagnosis of rotor faults in non-stationary conditions forms the third and most important part of this research. Several signal processing techniques are reviewed to analyze non-stationary signals. Three new algorithms are proposed that can track and detect rotor faults in non-stationary or transient current signals.

Inverse Synthetic Aperture Radar coherently processes the return signal from the target in order to construct the image of the target. The conventional methodology used for obtaining the image is the Fourier transform which is not capable of suppressing the Doppler change in the return signal. As a result, Range-Doppler image is degraded. A proper time-frequency transform suppresses the degradation due to time varying Doppler shift. In this thesis, high resolution joint-time frequency transformations that can be used in place of the conventional method are evaluated. Wigner-Ville Distribution, Adaptive Gabor Representation with Coarse-to-Fine search algorithm, and Time-Frequency Distribution Series are examined for the target imaging system. The techniques applied to sample signals compared with each other. The computational and memorial complexity of the methods are evaluated and compared to each other and possible improvements are discussed. The application of these techniques in the target imaging system is also performed and resulting images compared to each other.

Brain Computer Interfaces (BCIs) are systems developed in order to control devices by using only brain signals. In BCI systems, different mental activities to be performed by the users are associated with different actions on the device to be controlled. Spelling Paradigm is a BCI application which aims to construct the words by finding letters using P300 signals recorded via channel electrodes attached to the diverse points of the scalp. Reducing the letter detection error rates and increasing the speed of letter detection are crucial for Spelling Paradigm. By this way, disabled people can express their needs more easily using this application. In this thesis, two different methods, Support Vector Machine (SVM) and AdaBoost, are used for classification in the analysis. Classification and Regression Trees is used as the weak classifier of the AdaBoost. Time-frequency domain characteristics of P300 evoked potentials are analyzed in addition to time domain characteristics. Wigner-Ville Distribution is used for transforming time domain signals into time-frequency domain. It is observed that classification results are better in time domain. Furthermore, optimum subset of channels that models P300 signals with minimum error rate is searched. A method that uses both SVM and AdaBoost is proposed to select channels. 12 channels are selected in time domain with this method. Also, effect of dimension reduction is analyzed using Principal Component Analysis (PCA) and AdaBoost methods.