Search

This paper proposes a sensorless technique for an induction machine Flywheel Energy Storage System (FESS) based on a non-linear observer integrated with a total least-squares speed estimator taking into consideration the IM (Induction Machine) saturation effects. The nonlinear observer is based on an original formulation of the dynamic model of the IM including the magnetic saturation, rearranged in a space-state form, after assuming as state variables the stator current and the rotor magnetizing current space-vectors in the stator reference frame. The choice of the observer gain has been made by the use of Lyapunov's method. The speed signal needed by the non-linear observer for the flux estimation is provided by the Total Least Square (TLS) based speed estimator, which itself takes into consideration the magnetic saturation for the speed estimation. The proposed nonlinear observer has been tested in numerical simulation and verified experimentally on a suitably developed test set-up, demonstrating its applicability as far as FESS is concerned.

This paper proposes a technological solution with a scroll compressor driven by a high performance induction machine drive for the air management of PEMFC (proton exchange membrane fuel cells) for automotive applications. The torque-speed characteristics of a real scroll compressor have been measured and then emulated by a brushless IMPM (internal mounted permanent magnets) machine controlled in torque. This emulated scroll compressor has been driven by a field oriented controlled (FOC) induction motor drive. The whole car model as well as the PEMFC have been implemented by software on the same DSP which implements the drive control algorithm; in this way the hardware-in-the-loop structure has been employed for emulating the behaviour of a real car including the electrical supply of the PEMFC. The whole system has been tested with a classic European driving cycle. Two experimental rigs have been setup, one for characterizing the scroll compressor and the other for emulating the entire vehicle. The experimental results have shown that the speed reference profile, obtained during a urban driving cycle, is followed correctly and that the tank-to-wheel efficiency remains at quite high values confirming the goodness of the proposed technological solution. The hardware-in-the-loop approach of the experimental rig allows tests to be performed to verify the operation in steady and transient states with low cost and reduced employment of the Fuel Cell stack, with resulting increased durability.

This paper proposes a speed sensorless technique for induction motor drives based on the retrieval and tracking of the rotor slot harmonics (RSH). The RSH related to the rotor speed is first extracted from the stator phase current signature by the adoption of two cascaded ADALINEs (ADAptive Linear Element), whose output is the estimated slot harmonic. Then, the frequency of this slot harmonic as well as the speed is estimated by using minor space analysis (MSA) EXIN neural networks, which work on-line to iteratively compute the frequency of the slot harmonics based on MUSIC spectrum estimation theory. Thanks to its sample-based learning and the reduced mean square frequency estimation error, the speed estimation is fast and accurate. The proposed sensorless technique has been experimentally tested on a suitably developed test set-up with a 2- kW induction motor drive. It has been verified that this algorithm can track the rotor speed rapidly and accurately in a very wide speed range, working from rated speed down to 1.3 % of it.

Real time pattern recognition applications often deal with high dimensional data, which require a data reduction step which is only performed offline. However, this loses the possibility of adaption to a changing environment. This is also true for other applications different from pattern recognition, like data visualization for input inspection. Only linear projections, like the principal component analysis, can work in real time by using iterative algorithms while all known nonlinear techniques cannot be implemented in such a way and actually always work on the whole database at each epoch. Among these nonlinear tools, the Curvilinear Component Analysis (CCA), which is a non-convex technique based on the preservation of the local distances into the lower dimensional space, plays an important role. This paper presents the online version of CCA. It inherits the same features of CCA, is adaptive in real time and tracks non-stationary high dimensional distributions. It is composed of neurons with two weights: one, pointing to the input space, quantizes the data distribution, and the other, pointing to the output space, represents the projection of the first weight. This on-line CCA has been conceived not only for the previously cited applications, but also as a basic tool for more complex supervised neural networks for modelling very complex high dimensional data. This algorithm is tested on 2-D and 3-D synthetic data and on an experimental database concerning the bearing faults of an electrical motor, with the goal of novelty (fault) detection.

This paper proposes a speed sensorless technique for high performance induction motor drives based on the retrieval and tracking of the rotor slot harmonic. First, two cascaded ADALINEs (Adaptive Linear Elements) are used to extract the rotor slot harmonic (RSH) from the stator phase current signature, acting as adaptive filters, respectively in configuration band and notch, whose output consists of the RSH. Second, the MCA EXIN neurons are used to extract the eigenvector corresponding to the minimum eigenvalue of the autocorrelation matrix, which is formed by the ADALINEs' output sequence. Then, the slot frequency is estimated by using Pisarenko's theory with this retrieved minimum eigenvector, and subsequently the speed of the motor is estimated. Compared to the original Pisarenko's method however, not only the proposed algorithm can work recursively sample by sample, but the computational complexity and mean square frequency estimation error are largely reduced. The proposed sensorless technique has been experimentally tested on a suitably developed test set-up with a 2-kW induction motor drive. It has been verified that this algorithm can track the rotor speed rapidly and accurately in a very wide speed range, working from rated speed down to 1.3 % of it.

This paper presents for the first time the analytical solution of the constrained minimization for the on‐line estimation of the electrical parameters of an induction motor. The method is fully described mathematically and its goodness is verified experimentally on a suitably set up test bench. This methodology permits the almost correct computation of all the so called K‐parameters, which is not always the case in current literature, thus resulting in the correct estimation of the electrical parameters.

The neural solids are novel neural networks devised for solving optimization problems. They are dual to Hopfield networks, but with a quartic energy function. These solids are open architectures, in the sense that different choices of the basic elements and interfacings solve different optimization problems. The basic element is the neural resonator (triangle for the three dimensional case), composed of resonant neurons underlying a self-organizing learning. This module is able to solve elementary optimization problems such as the search for the nearest orthonormal matrix to a given one. Then, an example of a more complex solid, the neural decomposer, whose architecture is composed of neural resonators and their mutual connections, is given. This solid can solve more complex optimization problems such as the decomposition of the essential matrix, which is a very important technique in computer vision.

The paper presents a neural approach for the parameter estimation of adaptive IIR filters for linear system identification. It is based on a novel neuron, the TLS EXIN neuron, capable of resolving the TLS problem present in this kind of estimation, where noisy errors affect not only the observation vector but also the data matrix. After a survey of other techniques for solving such parameters estimations, the TLS EXIN neuron is compared both theoretically and numerically with the former techniques, resulting in improved performance. Moreover, it is also proved that the TLS EXIN neuron permits some powerful acceleration techniques, unlike the other approaches. These results are also shown numerically.

Condition monitoring schemes, able to deal with different sources of fault are, nowadays, required by the industrial sector to improve their manufacturing control systems. Pattern recognition approaches, allow the identification of multiple system's scenarios by means the relations between numerical features. The numerical features are calculated from acquired physical magnitudes, in order to characterize its behavior. However, only a reduced set of numerical features are used in order to avoid computational performance limitations of the artificial intelligence techniques. In this sense, feature reduction techniques are applied. Classical approaches analyze the features significance from a global data discrimination point of view. This paper, however, proposes a novel and reliable methodology to exploit the information contained in the original features set, by means a dedicated hierarchy of neural networks.

A parameter estimation method has been developed by manipulation of the dynamical equations describing the equivalent circuit of a 2-branch Double-Layer-Capacitor (DLC) supercapacitor model. This results in an over-determined matrix equation which can be solved by a least-squares method, in particular the (Total Least Squares) TLS EXIN neuron, making it exploitable also for on-line applications. Three parameters of the circuit can be computed in this way. The remaining parameters can be easily computed by two discharge tests, respectively one at constant current and the other at constant current load This method is quick, it needs only one set of measurement data and is robust to noise and stochastic measurement errors. Both simulation and experimental tests have been made to assess the methodology.

This paper deals with a simple, computationally not demanding, low cost scalar sensorless control of induction motor drives. The speed estimation has been performed on the basis of a PLL (Phase Locked Loop) algorithm, whose centre bandwith is tuned on-line on the basis of the reference values of the supply and slip frequencies provided to the PWM. This results in the PLL to track properly the machine speed in a very wide speed range. The proposed approach has been tested experimentally on a test set-up. Results show a good behaviour of the scalar sensorless drive in a very wide speed range from rated speed down to as much as 6% of the rated speed.

Mostly the faults in electrical machines are related with the bearings. Thus, a reliable bearing condition monitoring scheme able to detect either local or distributed defects are mandatory to avoid a breakdown in the machine. So far, the research has been carried out mainly in the detection of local faults, such as balls and raceways faults, but surface roughness is not so reported. This paper deals with a novel and reliable scheme capable to detect any fault that may occur in a bearing, based on EXIN Curvilinear Component Analysis, CCA, and Neural Network. The EXIN CCA, which is an improvement of the Curvilinear Component Analysis, has been conceived for data visualization, interpretation and classification for real time industrial applications. The effectiveness of this condition monitoring scheme has been verified by experimental results obtained from different operation conditions.

Bearing faults are the commonest form of malfunction associated with electrical machines. So far, the research has been carried out mainly in the detection of localized faults, but the diagnosis of distributed faults is still under development. In this context, this work presents a new scheme for detecting and classifying both kinds of faults. This work deals with a new diagnosis monitoring scheme, which is based on statistical-time features calculated from vibration signal, curvilinear component analysis for compression and visualization of the features behavior and a hierarchical neural network structure for classification. The obtained results from different operation conditions validate the effectiveness and feasibility of the proposed methodology.

EXIN CCA is an extension of the Curvilinear Component Analysis (CCA), which solves for the noninvariant CCA projection and allows representing data drawn under different operating conditions. It can be applied to data visualization, interpretation (as a kind of sensor of the underlying physical phenomenon) and classification for real time industrial applications. Here an example is given for bearing fault diagnostics in an electromechanical device.

A neural network based diagnosis system is presented for fully controlled converters. The system. by monitoring the current harmonics. is able to recognise the fault conditions and represents them in a 2D and 3D space. After describing the VQP algorithm a case study regarding three-phase converters is presented. By use of the VQP algorithm two different representations of normal and fault conditions have been obtained. The benefits of such approach are then discussed.

Detection and identification of induction machine faults through the stator current signal using higher order spectra analysis is presented. This paper proposes two higher order spectra techniques, namely the power spectrum and the slices of bi-spectrum used for the analysis of induction machine stator current leading to the detection of electrical failures within the rotor. Experimental signals have been analyzed highlighting that bi-spectrum results show their superiority in the accurate detection of rotor broken bars. Even when the induction machine is rotating at a low level of shaft load, the rotor fault detection is efficient.

This chapter contains sections titled: Introduction Generalized Total Least Squares (GeTLS EXIN) Approach GeTLS Stability Analysis Neural Nongeneric Unidimensional TLS Scheduling Accelerated MCA EXIN Neuron (MCA EXIN+) Further Considerations Simulations for the GeTLS EXIN Neuron ]]>

This chapter contains sections titled: GeMCA Approach Analysis of Matrix K Analysis of the Derivative of the Eigensystem of GeTLS EXIN Rank One Analysis Around the TLS Solution GeMCA spectra Qualitative Analysis of the Critical Points of the GeMCA EXIN Error Function Conclusions

This chapter contains sections titled: High-Order MCA Neurons Robust MCA EXIN Nonlinear Neuron (NMCA EXIN) Extensions of the Neural MCA ]]>

This chapter contains sections titled: Introduction Some TLS Applications Preliminaries Ordinary Least Squares Problems Basic TLS Problem Multidimensional TLS Problem Nongeneric Unidimensional TLS Problem Mixed OLS-TLS Problem Algebraic Comparisons Between TLS and OLS Statistical Properties and Validity Basic Data Least Squares Problem Partial TLS Algorithm Iterative Computation Methods Rayleigh Quotient Minimization Nonneural and Neural Methods ]]>

This paper presents a new linear neuron, the GeTLS EXIN neuron for the on-line estimation of the electrical parameters of the induction motor. This estimation problem is a non generic one, particularly for the flatness of the surface error along the K 2 direction, and current OLS neurons are not capable of facing it. Only the TLS EXIN neuron and the GeTLS EXIN neuron can cope with this issue for on-line applications. However, unlike the TLS EXIN neuron, the GeTLS EXIN neuron with its variation of the parameter ς, varies the landscape of the error surface, which then varies at each batch of data. This causes the surface to crumple up and reduce the flatness of the critical directions (K 2 in this case) with resulting improved accuracy.

Systems with high nonlinearities are, in general, very difficult to model. This is particularly true in geostatistics, where the problem of the estimation of a regionalized variable (RV) given only a small amount of measurement stations and a complex terrain surface is very challenging. This paper introduces a novel strategy, which couples the Curvilinear Component Analysis (CCA) and the Generalized Mapping Regressor (GMR). CCA, which is a nonlinear projector of a data manifold, is here used in order to find the intrinsic dimension of the data manifold, just giving an insight on the nonlinearities of the problem. This analysis drives the pre-processing of the data set used for the training phase of GMR. GMR is an incremental, self-organizing neural network which is able to model nonlinear functions by transforming the approximation into a pattern recognition problem. The presented approach is tested on the spatial estimation of the wind speed over the complex terrain of the isle of Sicily, in Italy. Wind speed maps resulting from this technique are presented and compared with a deterministic interpolator, the Inverse Distance Weighting (IDW) method.

This paper deals with the air management subsystem in fuel cell systems (FCS). It proposes to use a feedforward control action for improving a classical PI control of the FCS for preventing oxygen starvation by using the control scheme suggested by Kawato. This paper therefore, proposes the application of a recent neural network scheme, the GMR (Generalized Mapping Regressor Mapping) to implement a static feedforward controller, which is obtained with the inversion of the FCS model. The idea, for the air management subsystem, is to use a feedforward for optimising a classical PI control of the FCS for preventing oxygen starvation. This control system has been implemented both in numerical simulation and experimentally adopting a properly devised FCS in which the FC stack is realized by a buck-converter (emulator), that is by using a hardware-in-the-loop experimental rig.

This paper develops a new space-vector dynamic model of the induction machine including stator and rotor slotting effects. With the proposed approach, the slotting effect is considered only in the stator and rotor leakage variation, as proposed in literature. The model has been rearranged into a space-state form useful for control applications, since it contains the information of the rotor position. This model works both with a low and a high frequency supply, differently from most models in literature including slotting effects for control applications which usually model only the high frequency behaviour of the machine. The model has been implemented in numerical simulation and verified experimentally on three motors with the same stator and three different rotor configurations. Experimental results confirm the model, and are consistent with the results in literature.

This paper summarizes the research activity of the authors in the field of sensorless control of induction machine drives based on new linear neural techniques. In particular it describes and compares three speed observers: the MCA EXIN + MRAS Observer, the MCA EXIN + Reduced Order Observer and the TLS Full-order Luenberger Adaptive Observer. Common to all of three observers is the on-line estimation of the speed by a new linear neural technique, which solves in a recursive way a Total Least-Squares problem: one of them employs the TLS EXIN neuron and the other two the MCA EXIN + neuron, which is an improvement of the former. The speed observers have been verified in numerical simulations and experimentally on a test setup and have been also compared experimentally with one another.

This paper presents a unifying optimization methodology to improve the electrical machine optimal design. A new simplified response surface is proposed to offer the precision of numerical models based on FEM simulations

This paper proposes a new sensorless technique for induction motor drives based on a hybrid MRAS-neural technique. This MRAS method is an improvement of an already developed neural MRAS based sensorless method. In this paper the open-loop integration in the reference model is performed by an improved adaptive neural integrator. The adaptive model is based on a more accurate discrete current model and is trained on-line by a generalized least squares technique, the MCA EXIN + neuron, in which a parameterized learning algorithm is used. As a consequence, the speed estimation presents an improved convergence with higher accuracy and shorter settling time, because of the better transient behaviour of the neuron. A test bench has been set up to verify the methodology experimentally and the results prove its goodness at very low speeds ( below 4rad/s) and in zero-speed operation.

This paper presents a new adaptive integrator to avoid the DC drift phenomena and the initial condition problem typical of open-loop integration methods. This adaptive neural integrator has been applied to an open-loop speed estimator used in a rotor-flux-oriented vector control of an induction machine drive. Simulation and experimental results show the improvement of this new integrator as for the dynamical performance of the drive and the speed accuracy estimation in the low speed region (around 70 rpm). A comparison is then made experimentally with a classical speed estimation using a LP (Low-Pass) filter for integrating.

The Generalized Mapping Regressor (GMR) neural network is able to solve for inverse problems even when multiple solutions are given. In this case, it does not only identify these solutions (even if infinite, e.g. contours), but also specifies to which branch of the underlying mapping it belongs. It is also able to model mapping with discontinuities. The basic idea is the transformation of the mapping problem in a pattern recognition problem in a higher dimensional space (where the function branches are represented by clusters). Training is given by a multiresolution quantization represented by a pool of neurons whose number is determined by the training set. Then, neurons are linked each other by using some kind of local principal component analysis (LPCA). This phase is the most important and original. Other techniques (e.g. SVM's, mixture-of-experts) could work a priori on the same problems, but are not able to understand automatically when to stop the data quantization. This linking phase can be viewed as a reconstruction phase in which the correct clusters are recovered. The production phase uses a Gaussian kernel interpolation technique. Some examples conclude the paper.

This paper presents the experimental application of a new identification methodology, based on a constrained minimization technique, for the estimation of both the magnetisation curve of an induction machine and the variation of its electrical parameters caused by the main flux saturation. The parameter estimation algorithm is based on the dynamic mathematical model of the induction motor in the stationary reference frame and exploits a constrained least- squares algorithm. The experimental results obtained on a test set-up confirm the capability of the parameter estimation algorithm to estimate the electrical parameters of the machine correctly under different magnetic excitations. Moreover the method requires no complicated voltage waveform for supplying the motor, unlike other methods, as well as no a priori knowledge of the name-plate data of the machine and can be easily applied even with only one auto-transformer.

The motor current signature analysis has been considered as a standard for both electrical and mechanical fault detection in three-phase induction machine. However, even if the spectrum analysis is well known and has been published intensively in the literature, the problem of automatic classification for fault detection is still open. The aim of this paper is to present a new neural network approach for fault and speed detection using the current spectrum analysis. After the presentation of an original database creation, the classification using a particular neural network is developed. This network is conceived in order to output the posterior probabilities of class membership of the input, which allow the estimation of the level of confidence of the classification. Then, the classical motor slip detection algorithm is verified and the classification is experimentally performed on a squirrel-cage three-phase induction machine. The efficiency of the iteration process is shown together with the confusion matrix for the current spectrum analysis in the proposed method.

This paper presents the solution for the on-line identification of an induction motor by using the total least-squares (TLS) algorithm. In particular the TLS EXIN neuron has been employed as it is the only algorithm which solves for the TLS problem in a simple recursive way. The results thus obtained are compared with those obtained with an OLS (ordinary least-square) algorithm to show the substantial equivalence of the two techniques if no noise is present in the measurements. This is the first step for facing up to the estimation of electrical parameters of induction machines in noisy environments (EMC problems), for example in industrial environments.

This paper describes an innovative control technique for a DC electric drive when adaptation is to be taken into account. After a short overview of nonlinearities typical of electric machines, a neural-adaptive method based on the identification of the inverse model, is proposed for the control of such drives. The neural network adopted is a kind of clustering network, called the progressive learning network. Finally, simulations results are shown.

The robust version of the MCA EXIN linear neuron is introduced in order to solve typical minor component problems as the total least squares fitting in presence of impulsive and colored noise environments or in presence of outliers, i.e. in nonoptimal conditions for the traditional approaches. Furthermore, an analysis of the divergence of the robust neurons is made. The simulations show the better features of the NMCA EXIN neuron w.r.t. the existing nonneural and neural approaches, even in the case of high Gaussian noise together with strong outliers. This allows the use of this neuron for some very difficult problems, like in computer vision, just giving the possibility of massively high parallel architectures.

This paper deals with the exploitation of the epipolar constraint information for the construction of the essential matrix (fundamental matrix for uncalibrated images), which, once decomposed, solves the structure from motion problem. This technique has been longly considered inferior to the techniques which use the optical flow information, because of its high sensitivity to noise. The approach used here, which is particularly robust to noise, both demonstrates the validity and presents an extension of the essential matrix approach. Once established the fact that the problem at hand is Total Least Squares (TLS) with a certain structure, a statistical analysis of the problem suggests the use of the Constrained TLS in order to take in account the linear dependences of the noise components in the coefficients. This leads to the CTLS and CTLSn EXIN neurons which are simple variants (change of metrics) of the TLS EXIN linear neuron which solves the basic TLS problem. These neurons are able to yield good results even in the presence of an outlier contamination of half the point matches.

In this paper a neural network based architecture, which combines supervised and unsupervised learning for the static security assessment of power systems, is presented. The proposed method allows the on-line security evaluation of a possible outage simply by considering the position of the neuron activated by the pre-fault state vector in an output map, allowing an easy and immediate view of the contingency risks. The mapping capabilities of two unsupervised neural networks, SOM (self-organising map) and CCA (curvilinear component analysis), are compared. Numerical tests, carried out on a study system, are presented and discussed.

This paper deals with the application of the General Mapping Regressor (GMR) neural network to the direct torque control DTC of an induction motor. In particular it shows that the GMR neural network is able to correctly learn the classical DTC, as well as any other more involved control strategy. A suitable test bench has been set up in order to verify the performance of the neural controller.

In this thesis, the neural approach TLS EXIN was proposed in a new way in order to estimate the position of a GPSreceiver. The general idea of this approach is to develop a more robust method for calculating the position.The ¿pseudorange alone¿ method is one of the simplest techniques and most widely used for estimating the GPSpositioning and it results in solving an overdetermined system of linear equations Ax¿b. In general, the ordinaryleast squares (OLS) and weighted least squares (WLS) are the commonly used methods to estimate the position ofa receiver for their quickness and robustness, but the particular structure of the data matrix A and the noise affectingits entries are not considered. This thesis, instead, aims to address these problems and study the behavior of leastsquares (LS) methods in the presence of a noisy data matrix A.The approach of total least squares (TLS) takes into account the noise in A and b. It is a less robust technique thanOLS and more sensitive to data changes. It is general solved as a direct method. Instead, the neural network TLSEXIN, which is an iterative algorithm (gradient flow) for solving TLS, works better both because it may exploit theinitial condition information derived from the previous epochs, and, in case of null initial conditions, yields anaccurate estimate even in case of close-to-degenerate data matrix.To perform tests between different methods of least squares, two sets of data were collected. The first one comesfrom the TERIA network and includes data collected from different reference stations located throughout France.While the second one is the result of a measurement campaign using GPS (Ublox NL-6002U).Using the real data, a low condition number has been estimated: in this case all LS methods yield equivalentestimates: as a consequence, OLS and WLS are to be preferred for their low computational cost. However, theworst scenario has been investigated which may occur in case of a distant satellite, resulting in the ill-conditioning ofthe GPS problem. This extreme situation justifies the use of the TLS EXIN neural network. The results obtainedconfirm of this approach even for very large condition numbers., Dans ce mémoire, une nouvelle approche neuronale TLS EXIN est proposée pour estimer la position d'un récepteurGPS. L'idée générale de cette approche est d'avoir une méthode plus robuste pour le calcul de la position.Le pseudorange alone est l'une des techniques les plus simples et les plus utilisées pour le positionnement GPS.Cette technique nécessite la résolution d'un système surdéterminé d'équations linéaires Ax+/-b. En général, lesmoindres carrés ordinaires (OLS) et les moindres carrés pondérés (WLS) sont les méthodes couramment utiliséespour estimer la position d'un récepteur grâce à leur rapidité et leur robustesse, mais la structure particulière de lamatrice de données A et les bruits affectant ses entrées ne sont pas considérés. Au contraire, cette thèse a pourobjectif d'analyser ces problèmes et d'étudier le comportement des méthodes des moindres carrés (LS) enprésence d'une matrice de données A bruitée.L'approche des moindres carrés totaux (TLS) prend en compte le bruit dans la matrice de données A ainsi que dansle vecteur d'observation b. Cette dernière est une technique moins robuste que OLS et plus sensible auchangement des données, elle est en général résolue par une méthode directe. La méthode TLS EXIN basée surles réseaux de neurones est un algorithme itératif (flux gradient) pour résoudre le problème TLS. Elle donne unmeilleur résultat parce qu'elle peut exploiter les informations d'état initial provenant des époques précédentes et, encas de conditions initiales nulles, donne une estimation précise même en cas de problème singulier.Pour réaliser des comparaisons entre les différentes méthodes des moindres carrés (LS), deux jeux de données ontété collectés. Le premier jeu de données est issu du réseau TERIA et comporte des données collectées depuisdifférentes stations de référence situées dans toute la France. Le deuxième jeu de données est le résultat d'unecampagne de mesures utilisant un appareil GPS (Ublox NL-6002U).Grâce à ces données réelles, un nombre de conditionnement bas a été estimé. Dans ce cas, toutes les méthodesLS donnent des estimations équivalentes, et le choix du meilleur algorithme (OLS, et surtout, WLS) est privilégiépour leur rapidité de calcul. Cependant, le pire scénario qui puisse se produire a été étudié (dans le cas d'unsatellite éloigné), et ont été observés des mauvais conditionnements du problème de GPS (nombre deconditionnement élevé). Cette situation extrême justifie l'utilisation du réseau neuronal TLS EXIN. Les résultatsobtenus confirment cette approche, même pour un nombre de conditionnement élevé.____________________________________________________________________________________________

The thesis first studies the relation between the rotor slot harmonics (RSHs) and the instan-taneous rotor speed. To directly track the RSH, the requirements of the system are fully ad-dressed.Second, the thesis presents a sensorless scheme based on phase-locked loop (PLL): The centre bandwidth is tuned on-line on the basis of the reference values of the supply and slip frequencies provided to the PWM converter, the PLL is tuned to track the machine rotor slot-ting harmonic without the need of any high frequency signal injection, neither rotating nor pulsating. This speed estimation scheme, which is suitable for the scalar controller, had been integrated with the scalar drive, leading to a simple, computationally not demanding, low cost sensorless IM drives. The experiment results show that the system is able to track the machine speed in a very wide speed range.Finally, an improved sensorless scheme based on minor component analysis (MCA) neu-rons is described. According to the Pisarenko’s theory, it has been verified that the MC which lies in the noise subspace is orthogonal to the signal subspace, thus, the signal frequencies contained in the input can be computed from a polynomial formed by the MC. Conventional-ly, this will require the bulky eigen-decomposition, nevertheless, the neural method proposed in this thesis can retrieve the MC recursively with less computation and improved error per-formance (the solution is of total least square meaning). Moreover, the speed estimator is ap-plied to the scalar drive with experimental verification, the overall system is well behaved, and the MCA method enhanced by neural networks has provided a good potential in the ap-plication of harmonics retrieve.; La thèse étudie tout d’abord la relation entre les harmoniques à fentes du rotor (RSHs) et la vitesse du rotor instantanée. Pour suivre directement l'RSH, les exigences du système sont pleinement prises en compte.Dans un deuxième temps, les travaux de thèse ont permis de développer un système sans capteur en fonction de boucle à verrouillage de phase (PLL): La largeur de bande centrale est réglée en ligne sur la base des valeurs de référence, des fréquences d'alimentation et de glissement prévues au convertisseur PWM, la PLL est réglée pour suivre le rotor de la machine à RSH sans la nécessité de toute injection de signal à haute fréquence, ni en rotation, ni de pulsation. Ce système d'estimation de vitesse, qui est approprié pour le contrôleur scalaire, avait été intégré avec le lecteur scalaire, conduisant à un simple calcul peu exigeant, à faible coût de l’entraînement de la machine à induction sans capteur à faible coût. Les résultats expérimentaux montrent que le système est en mesure de suivre la vitesse de la machine dans une plage de vitesse très étendue.Enfin, un système sans capteur amélioré basé sur l'analyse de composant mineur (MCA) neurones est décrit. Selon la théorie de Pisarenko, il a été vérifié que le MC qui se trouve dans le sous-espace de bruit est orthogonale au sous-espace de signal, par conséquent, les fré-quences de signal contenues dans l'entrée peuvent être calculées à partir d'un polynôme formé par la MC. Classiquement, ce qui nécessitera la décomposition propre encombrants, néan-moins, la méthode de neurones proposée dans cette thèse peut récupérer le MC de façon ré-cursive avec moins de calculs et des performances améliorées d'erreur (la solution est sur un total de moins sens carré). En outre, l'estimateur de vitesse est appliquée à l'entraînement scalaire avec vérification expérimentale, l'ensemble du système se comporte bien, et la méthode MCA renforcée par réseaux neuronaux a fourni un bon potentiel dans l'application des harmoniques récupérer.

The thesis first studies the relation between the rotor slot harmonics (RSHs) and the instan-taneous rotor speed. To directly track the RSH, the requirements of the system are fully ad-dressed.Second, the thesis presents a sensorless scheme based on phase-locked loop (PLL): The centre bandwidth is tuned on-line on the basis of the reference values of the supply and slip frequencies provided to the PWM converter, the PLL is tuned to track the machine rotor slot-ting harmonic without the need of any high frequency signal injection, neither rotating nor pulsating. This speed estimation scheme, which is suitable for the scalar controller, had been integrated with the scalar drive, leading to a simple, computationally not demanding, low cost sensorless IM drives. The experiment results show that the system is able to track the machine speed in a very wide speed range.Finally, an improved sensorless scheme based on minor component analysis (MCA) neu-rons is described. According to the Pisarenko’s theory, it has been verified that the MC which lies in the noise subspace is orthogonal to the signal subspace, thus, the signal frequencies contained in the input can be computed from a polynomial formed by the MC. Conventional-ly, this will require the bulky eigen-decomposition, nevertheless, the neural method proposed in this thesis can retrieve the MC recursively with less computation and improved error per-formance (the solution is of total least square meaning). Moreover, the speed estimator is ap-plied to the scalar drive with experimental verification, the overall system is well behaved, and the MCA method enhanced by neural networks has provided a good potential in the ap-plication of harmonics retrieve., La thèse étudie tout d’abord la relation entre les harmoniques à fentes du rotor (RSHs) et la vitesse du rotor instantanée. Pour suivre directement l'RSH, les exigences du système sont pleinement prises en compte.Dans un deuxième temps, les travaux de thèse ont permis de développer un système sans capteur en fonction de boucle à verrouillage de phase (PLL): La largeur de bande centrale est réglée en ligne sur la base des valeurs de référence, des fréquences d'alimentation et de glissement prévues au convertisseur PWM, la PLL est réglée pour suivre le rotor de la machine à RSH sans la nécessité de toute injection de signal à haute fréquence, ni en rotation, ni de pulsation. Ce système d'estimation de vitesse, qui est approprié pour le contrôleur scalaire, avait été intégré avec le lecteur scalaire, conduisant à un simple calcul peu exigeant, à faible coût de l’entraînement de la machine à induction sans capteur à faible coût. Les résultats expérimentaux montrent que le système est en mesure de suivre la vitesse de la machine dans une plage de vitesse très étendue.Enfin, un système sans capteur amélioré basé sur l'analyse de composant mineur (MCA) neurones est décrit. Selon la théorie de Pisarenko, il a été vérifié que le MC qui se trouve dans le sous-espace de bruit est orthogonale au sous-espace de signal, par conséquent, les fré-quences de signal contenues dans l'entrée peuvent être calculées à partir d'un polynôme formé par la MC. Classiquement, ce qui nécessitera la décomposition propre encombrants, néan-moins, la méthode de neurones proposée dans cette thèse peut récupérer le MC de façon ré-cursive avec moins de calculs et des performances améliorées d'erreur (la solution est sur un total de moins sens carré). En outre, l'estimateur de vitesse est appliquée à l'entraînement scalaire avec vérification expérimentale, l'ensemble du système se comporte bien, et la méthode MCA renforcée par réseaux neuronaux a fourni un bon potentiel dans l'application des harmoniques récupérer.