Abstract This thesis presents that the closed-loop sensitivity function shaping combined with -synthesis applies to design the controller with structured uncertainties. The sensitivity function shaping is directly based on the indices of the closed-loop performances. The closed-loop frequency response and the robust stability for the system could approach the designed performances by adjusting weighting functions. Since the robust performance of the closed-loop systems bases on the index of the open-loop function in the loop shaping, it may not accomplish the requirement of the designer. The loop shaping can¡¦t be applied to design controllers for the system with structured uncertainties. Therefore, using the closed-loop sensitivity function shaping to design controller will contain the system with structured uncertainties and satisfy the closed-loop performance.

The Rapid Thermal Chemical Vapor Deposition (RTCVD) system is an emerging and promising technology in semiconductor manufacturing which possess advantages of rapidly increasing wafer temperature and reducing the thermal budget over traditional batch processing. In recent years, the growth of thin films in the manufacture of semiconductor devices has been widely employed in the industry. Because the influences of processing variables on RTCVD systems may lead to spatial wafer temperature non-uniformity, the precise control of wafer temperature is an important issue up to the present. In this paper the complementary sensitivity function shaping based on H-infinite control theory is applied to design robust controllers for the single-input/single-output (SISO) model of the RTCVD system, the multi-input/multi-output (MIMO) model of the RTCVD system, and the MIMO model with multiplicative uncertainties. Through control the power of the tungsten-halogen lamps, it can achieve the temperature tracking with good uniformity. Finally, the computer simulation results are obviously that the performance of the proposed controllers is satisfactory.

Η ευαισθησία των συστημάτων είναι ένα σημαντικό θέμα της Θεωρίας Συστημάτων, το οποίο καλύπτεται σε εισαγωγικό επίπεδο με αυτή την εργασία. Η ευαισθησία αφορά τις διαφορές ανάμεσα στο πραγματικό σύστημα και στο μαθηματικό μοντέλο: αν κάποιες παράμετροι διαφέρουν αρκετά ανάμεσα στο πραγματικό σύστημα και στο μαθηματικό μοντέλο και η συμπεριφορά του συστήματος εξαρτάται κατά μεγάλο βαθμό από αυτές τις παραμέτρους, τότε η χρησιμότητα του μαθηματικού μοντέλου θα είναι πολύ μικρή αν δεν γνωρίζουμε ταυτόχρονα την παραμετρική ευαισθησία του συστήματος, δηλαδή την επίδραση των μεταβολών των παραμέτρων πάνω στην δυναμική συμπεριφορά του συστήματος. Βασικό ρόλο παίζει η προσομοίωση των εξισώσεων ευαισθησίας κατάστασης. Καθορίζουμε αρχικά τις εξισώσεις ευαισθησίας κατάστασης για παραμέτρους τύπου α και δείχνουμε ότι οι συναρτήσεις ευαισθησίας κατάστασης ενός συνεχούς συστήματος με χρονικά αμετάβλητες παραμέτρους καθορίζονται πάντα από ένα γραμμικό σύστημα διαφορικών εξισώσεων με μηδενικές αρχικές συνθήκες. Στην συνέχεια θα επεκταθούμε και στην εύρεση των συναρτήσεων ευαισθησίας κατάστασης για την περίπτωση που έχουμε παραμέτρους τύπου β και θα δούμε ότι και εδώ οι εξισώσεις ευαισθησίας κατάστασης είναι πάντα γραμμικές και οι αρχικές συνθήκες είναι ή μηδέν ή μονάδα. Στη συνέχεια προχωράμε στην εύρεση των συναρτήσεων ευαισθησίας κατάστασης για την περίπτωση που έχουμε παραμέτρους τύπου λ, αφού δούμε πρώτα το πότε μπορούμε να χρησιμοποιήσουμε το μειωμένο ονομαστικό μοντέλο. Επί πλέον, εξετάζουμε εν συντομία και τον καθορισμό των εξισώσεων ευαισθησίας εξόδου. Για την περίπτωση που έχουμε να μελετήσουμε και να συγκρίνουμε την παραμετρική ευαισθησία συστημάτων ανοικτού και κλειστού βρόχου καθώς επίσης και στην σύνθεση συστημάτων "αναίσθητων" σε παραμετρικές μεταβολές, είναι αναγκαίο να έχουμε έναν ορισμό ευαισθησίας που να είναι ανεξάρτητος από την μορφή του σήματος εισόδου, αλλά να εξαρτάται μόνο από την δομή του συστήματος. Αυτή η απαίτηση ικανοποιείται με τους ορισμούς ευαισθησίας στο πεδίο συχνότητας, οι οποίοι βασίζονται στην συνάρτηση μεταφοράς ή στον πίνακα μεταφοράς του συστήματος. Εξετάζουμε τη συνάρτηση ευαισθησίας του Bode, τη συνάρτηση ευαισθησίας του Horowitz και τη συγκριτική συνάρτηση ευαισθησίας των Perkins και Cruz. Η τελευταία μπορεί να γενικευθεί και σε μη γραμμικά, χρονικά μεταβαλλόμενα συστήματα. / The sensitivity of systems is an important theme of Systems Theory, which is covered to an introductory level in this work. The sensitivity concerns differences between the real system and the mathematical model: if some parameters are quite different between the real system and the mathematical model and the behavior of the system depends strongly on these parameters, the utility of the mathematical model will be very little if we do not know both the parametric sensitivity of the system, i.e. the effect of changes in parameters on the dynamic behavior of the system. A key role is played by simulating the state sensitivity equations. First we determine the state of sensitivity equations for parameters of type a and show that the state of sensitivity functions of a continuous system with time invariant parameters are always determined by a linear system of differential equations with zero initial conditions. Then we expand and find the state sensitivity functions for the case where we have parameters of type b and we see that here the situation sensitivity equations are always linear and the initial conditions are zero. Then we move on finding state sensitivity functions for the type lambda parameters, after we look when the reduced nominal model can use. Moreover, we look briefly at the output sensitivity equations. In case we have to study and compare the parametric sensitivity of open and closed-loop systems as well as in the synthesis systems "unconscious" to parametric changes, it is necessary to have a definition of sensitivity that is independent of the input signal format, but can depend only on the structure of the system. This requirement is satisfied by the definitions of sensitivity in the frequency domain, which are based on the transfer function or the system operator panel. We examine the sensitivity function of Bode, the sensitivity function of Horowitz and the comparative context sensitivity of Perkins and Cruz. The latter can be generalized to nonlinear, time-varying systems.

A crescente demanda pela melhoria operacional dos processos aliada ao desenvolvimento da tecnologia da informação tornam a utilização de controladores preditivos baseados em modelos (MPC) uma prática comum na indústria. Estes controladores estimam, a partir dos dados de planta e de um modelo do processo, uma sequência de ações de controle que levam as variáveis ao valor desejado de forma otimizada. Dessa forma, dentre os parâmetros de configuração de um MPC, a baixa qualidade do modelo é, indiscutivelmente, a mais importante fonte de degradação de seu desempenho. Este trabalho propõe uma série de metodologias para a avaliação da qualidade do modelo do controlador preditivo, as quais consideram sua velocidade em malha fechada. Tais metodologias são baseadas na filtragem dos erros de simulação a partir função nominal de sensibilidade, e possuem a capacidade de informar o impacto dos problemas de modelagem no desempenho do sistema, além de localizar as variáveis controladas que estão com tais problemas e se os mesmos são provenientes de uma discrepância no modelo ou de um distúrbio não medido. As técnicas ainda possuem a vantagem de serem independentes do setpoint, o que as torna flexível de também serem utilizadas em controladores nos quais as variáveis são controladas por faixas. A abordagem proposta foi testada em dois estudos de caso simulados, sendo eles: a Fracionadora de Óleo Pesado da Shell e a Planta de Quatro tanques Cilíndricos. As técnicas também foram avaliadas em dados de processo da Unidade de Coqueamento Retardado de uma refinaria. Os resultados indicam que as mesmas apresentam resultados coerentes, corroborando seu elevado potencial de aplicação industrial. / The growing demand for operational improvement and the development of information technology make the use of model predictive controllers (MPCs) a common practice in industry. This kind of controller uses past plant data and a process model to estimate a sequence of control actions to lead the variables to a desired value following an optimal policy. Thus, the model quality is the most important source of MPC performance degradation. This work proposes a series of methods to investigate the controller model quality taking into account its closed loop performance. The methods are based on filtering the simulation errors using the nominal sensitivity function. They are capable detect the impact of modeling problems in the controller performance, and also to locate the controlled variables that have such problems and if it is caused by a model-plant mismatch or unmeasured disturbance. The techniques have the advantage to be setpoint independent, making them flexible to be also used in MPCs with controlled variables working by range. The proposed approach was tested in two simulated case studies The Shell Heavy Oil Fractionator Process and The Quadruple-tanks Process. The methods are also evaluated in process data of the Delayed Coking Unit of a Brazilian refinery. Results indicate that the method is technically coherent and has high potential of industrial application.

A crescente demanda pela melhoria operacional dos processos aliada ao desenvolvimento da tecnologia da informação tornam a utilização de controladores preditivos baseados em modelos (MPC) uma prática comum na indústria. Estes controladores estimam, a partir dos dados de planta e de um modelo do processo, uma sequência de ações de controle que levam as variáveis ao valor desejado de forma otimizada. Dessa forma, dentre os parâmetros de configuração de um MPC, a baixa qualidade do modelo é, indiscutivelmente, a mais importante fonte de degradação de seu desempenho. Este trabalho propõe uma série de metodologias para a avaliação da qualidade do modelo do controlador preditivo, as quais consideram sua velocidade em malha fechada. Tais metodologias são baseadas na filtragem dos erros de simulação a partir função nominal de sensibilidade, e possuem a capacidade de informar o impacto dos problemas de modelagem no desempenho do sistema, além de localizar as variáveis controladas que estão com tais problemas e se os mesmos são provenientes de uma discrepância no modelo ou de um distúrbio não medido. As técnicas ainda possuem a vantagem de serem independentes do setpoint, o que as torna flexível de também serem utilizadas em controladores nos quais as variáveis são controladas por faixas. A abordagem proposta foi testada em dois estudos de caso simulados, sendo eles: a Fracionadora de Óleo Pesado da Shell e a Planta de Quatro tanques Cilíndricos. As técnicas também foram avaliadas em dados de processo da Unidade de Coqueamento Retardado de uma refinaria. Os resultados indicam que as mesmas apresentam resultados coerentes, corroborando seu elevado potencial de aplicação industrial. / The growing demand for operational improvement and the development of information technology make the use of model predictive controllers (MPCs) a common practice in industry. This kind of controller uses past plant data and a process model to estimate a sequence of control actions to lead the variables to a desired value following an optimal policy. Thus, the model quality is the most important source of MPC performance degradation. This work proposes a series of methods to investigate the controller model quality taking into account its closed loop performance. The methods are based on filtering the simulation errors using the nominal sensitivity function. They are capable detect the impact of modeling problems in the controller performance, and also to locate the controlled variables that have such problems and if it is caused by a model-plant mismatch or unmeasured disturbance. The techniques have the advantage to be setpoint independent, making them flexible to be also used in MPCs with controlled variables working by range. The proposed approach was tested in two simulated case studies The Shell Heavy Oil Fractionator Process and The Quadruple-tanks Process. The methods are also evaluated in process data of the Delayed Coking Unit of a Brazilian refinery. Results indicate that the method is technically coherent and has high potential of industrial application.

A crescente demanda pela melhoria operacional dos processos aliada ao desenvolvimento da tecnologia da informação tornam a utilização de controladores preditivos baseados em modelos (MPC) uma prática comum na indústria. Estes controladores estimam, a partir dos dados de planta e de um modelo do processo, uma sequência de ações de controle que levam as variáveis ao valor desejado de forma otimizada. Dessa forma, dentre os parâmetros de configuração de um MPC, a baixa qualidade do modelo é, indiscutivelmente, a mais importante fonte de degradação de seu desempenho. Este trabalho propõe uma série de metodologias para a avaliação da qualidade do modelo do controlador preditivo, as quais consideram sua velocidade em malha fechada. Tais metodologias são baseadas na filtragem dos erros de simulação a partir função nominal de sensibilidade, e possuem a capacidade de informar o impacto dos problemas de modelagem no desempenho do sistema, além de localizar as variáveis controladas que estão com tais problemas e se os mesmos são provenientes de uma discrepância no modelo ou de um distúrbio não medido. As técnicas ainda possuem a vantagem de serem independentes do setpoint, o que as torna flexível de também serem utilizadas em controladores nos quais as variáveis são controladas por faixas. A abordagem proposta foi testada em dois estudos de caso simulados, sendo eles: a Fracionadora de Óleo Pesado da Shell e a Planta de Quatro tanques Cilíndricos. As técnicas também foram avaliadas em dados de processo da Unidade de Coqueamento Retardado de uma refinaria. Os resultados indicam que as mesmas apresentam resultados coerentes, corroborando seu elevado potencial de aplicação industrial. / The growing demand for operational improvement and the development of information technology make the use of model predictive controllers (MPCs) a common practice in industry. This kind of controller uses past plant data and a process model to estimate a sequence of control actions to lead the variables to a desired value following an optimal policy. Thus, the model quality is the most important source of MPC performance degradation. This work proposes a series of methods to investigate the controller model quality taking into account its closed loop performance. The methods are based on filtering the simulation errors using the nominal sensitivity function. They are capable detect the impact of modeling problems in the controller performance, and also to locate the controlled variables that have such problems and if it is caused by a model-plant mismatch or unmeasured disturbance. The techniques have the advantage to be setpoint independent, making them flexible to be also used in MPCs with controlled variables working by range. The proposed approach was tested in two simulated case studies The Shell Heavy Oil Fractionator Process and The Quadruple-tanks Process. The methods are also evaluated in process data of the Delayed Coking Unit of a Brazilian refinery. Results indicate that the method is technically coherent and has high potential of industrial application.

La caractérisation expérimentale de la réponse vibro-acoustique de panneaux excités par des champs de pression aléatoires est d'un grand intérêt autant en recherche que pour des applications industrielles. Deux excitations sont particulièrement étudiées: le champ acoustique diffus les pressions fluctuantes induites par une couche limite turbulente. Les moyens d’essais associés à ces excitations (chambre réverbérante, soufflerie, essais in situ) peuvent être très coûteux, difficilement contrôlables et rarement comparables entre laboratoires ou centre d’essais (installations de différentes dimensions ou mises en œuvres). La reproductibilité des mesures peut alors être remise en cause, ce qui rend difficile la comparaison entre différentes solutions technologiques. Il y a donc un fort intérêt à disposer d'un outil de laboratoire permettant de reproduire l’effet de ces excitations aléatoires dans un environnement contrôlé. Dans ce contexte, cette thèse propose de développer une méthode expérimentale permettant de caractériser le comportement vibro-acoustique de panneaux sous un champ de pression aléatoire en s’affranchissant des moyens d’essais usuels. Les approches que l’on étudie se basent sur la formulation mathématique du problème dans le domaine des nombres d'onde. Celle-ci met en évidence une séparation explicite des contributions de l'excitation (via l’interspectre de pression pariétale), de celles du comportement vibro-acoustique du panneau (via les fonctions de sensibilité). A partir de la connaissance de l'interspectre de pression pariétale, il suffit alors de déterminer expérimentalement les fonctions de sensibilité du panneau afin de déterminer par post-traitement sa réponse à l'excitation considérée. Deux méthodes permettant de déterminer les fonctions de sensibilité sont étudiées numériquement et validées expérimentalement : la méthode dite de l’antenne synthétique et une méthode basée sur des principes de réciprocité. Pour étudier la validité de ces méthodes, on compare leurs résultats à ceux obtenus par des moyens standards pour deux types de panneaux et les deux types d’excitations évoqués précédemment. / Abstract: The experimental vibro-acoustic characterization of panels submitted to random pressure fields is of great interest in the industry as well as in research laboratories. For the transport sector, this type of excitation can be found when a turbulent flow develops at the wall of a moving vehicle for example. The pressure fluctuations induced by the turbulent boundary layer excite the panels which radiate a noise inside the cabin. The experimental reproduction of those pressure fluctuations requires test means which can be very costly (i.e., wind tunnel, in situ tests) and whose physical parameters can hardly be controlled. The repeatability of measurements can thereby be questioned which makes it hard to compare different technological solutions. A second example of random pressure field is the diffuse acoustic field. This latter is usually reproduced in a reverberant room which is often coupled with an anechoic chamber by means of the panel whose acoustic insulation is to be tested. A pressure field is assumed to be diffuse if the acoustic energy comes from every direction with an equiprobable intensity of the incident waves. This assumption is never fully reached in practice (lack of grazing incident waves, strong modal behavior of the room at low frequencies, etc.). A laboratory tool which allows reproducing the effect of those random excitations in a controlled environment is therefore of great interest. In this context, this thesis aims at developing an experimental method to characterize the vibro-acoustic behavior of panels under random pressure fields without using the common test means (wind tunnel, reverberant room, in situ tests, etc.). For relevance sake, this approach must compensate for the previously stated issues. The approaches studied in this work are based on the mathematical formulation of the problem in the wavenumber domain. This latter allows an explicit separation of the contributions of the excitation via the wall-pressure cross-spectrum, from those of the vibro-acoustic behavior of the panel via so-called ‘sensitivity functions’. Assuming the wall-pressure cross-spectrum of the excitation is known, it is only required to experimentally determine those sensitivity functions, on the panel or in the acoustic medium, to determine the response of the panel to the considered excitation by post-processing. Two methods aiming at determining the sensitivity functions will be numerically and experimentally studied: the source scanning technique and the method based on the reciprocity principle. Results obtained with those method are compared to measurements using standard test means to attest the validity of those methods. Several vibro-acoustic indicators will be confronted while considering the two previously mentioned excitations and for two types of panels: an academic panel and a ‘complex’ from the aeronautic sector. This latter shows the applicability of the method in an industrial context.

Image sizes have increased exponentially in recent years. The resulting high-resolution images are often viewed via remote image browsing. Zooming and panning are desirable features in this context, which result in disparate spatial regions of an image being displayed at a variety of ( spatial) resolutions. When an image is displayed at a reduced resolution, the quantization step sizes needed for visually lossless quality generally increase. This paper investigates the quantization step sizes needed for visually lossless display as a function of resolution, and proposes a method that effectively incorporates the resulting ( multiple) quantization step sizes into a single JPEG 2000 codestream. This codestream is JPEG 2000 Part 1 compliant and allows for visually lossless decoding at all resolutions natively supported by the wavelet transform as well as arbitrary intermediate resolutions, using only a fraction of the full-resolution codestream. When images are browsed remotely using the JPEG 2000 Interactive Protocol ( JPIP), the required bandwidth is significantly reduced, as demonstrated by extensive experimental results.

The work describes the development of an adjustment algorithm that will replace today’s manual adjustment of Lead- and Lag-filters at ABB Robotics industrial robots. The adjustment is mainly for the filter parameters in the control system. Three different algorithms have been created to replace the manual adjustment. One of them is based on the criterion from the manual adjustment and another one on optimization of the sensitivity function. The last one is a combination of the two. The report describe the development of the algorithms and how they behave during simulations and tests on real robots. All three algorithms are several times faster then todays adjustment and present equivalent results at the same time. / Arbetet beskriver framtagandet av trimningsalgoritmer som ska överta dagens manuella trimning av Lead- och Lag-filter på ABB Robotics industrirobotar. Trimningen riktar sig framförallt till filterparametrarna i styrsystemet. Tre olika algoritmer har tagits fram som arvtagare till dagens manuella trimning. En av algoritmerna utgår från kriterier i dagens trimning medan en annan använder sig av optimering av känslighetsfunktionen. Den sista kan ses som en blandning av de två algoritmerna. I rapporten kan du följa algoritmernas tillkomst, hur dess resultat beter sig vid simuleringstester samt verkliga robottester. Alla tre algoritmerna är mångfaldigt snabbare än dagens trimning och presenterar samtidigt likvärdiga resultat.

Image quality measures are used to optimize image processing algorithms and evaluate their performances. The only reliable way to assess image quality is subjective evaluation by human observers, where the mean value of their scores is used as the quality measure. This is known as mean opinion score (MOS). In addition to this measure there are various objective (quantitative) measures. Most widely used quantitative measures are: mean squared error (MSE), peak signal to noise ratio (PSNR) and signal to noise ratio (SNR). Since these simple measures do not always produce results that are in agreement with subjective evaluation, many other quality measures have been proposed. They are mostly various modifications of MSE, which try to take into account some properties of human visual system (HVS) such as nonlinear character of brightness perception, contrast sensitivity function (CSF) and texture masking. In these approaches quality measure is computed as MSE of input image intensities or frequency domain coefficients obtained after some transform (DFT, DCT etc.), weighted by some coefficients which account for the mentioned properties of HVS. These measures have some advantages over MSE, but their ability to predict image quality is still limited. A different approach is presented here. Quality measure proposed here uses simple model of HVS, which has one user-defined parameter, whose value depends on the reference image. This quality measure is based on the average value of locally computed correlation coefficients. This takes into account structural similarity between original and distorted images, which cannot be measured by MSE or any kind of weighted MSE. The proposed measure also differentiates between random and signal dependant distortion, because these two have different effect on human observer. This is achieved by computing the average correlation coefficient between reference image and error image. Performance of the proposed quality measure is illustrated by examples involving images with different types of degradation.