Résumé : Le travail réalisé au cours de cette thèse traite de la mise au point de méthodes d’estimation d’état robuste, avec deux domaines d’application en ligne de mire. Le premier concerne le positionnement sécuritaire en transport. L’objectif est de fournir la position et la vitesse du véhicule sous la forme d’intervalles avec un grand degré de confiance. Le second concerne la synthèse de capteurs logiciels pour les bioprocédés, et en particulier la reconstruction des concentrations de composants réactionnels à partir d’un nombre limité de mesures et d’un modèle mathématique interprétant le comportement dynamique de ces composants. L’objectif principal est de concevoir des algorithmes qui puissent fournir des estimations acceptables en dépit des incertitudes provenant de la mauvaise connaissance du système comme les incertitudes sur les paramètres du modèle ou les incertitudes de mesures. Dans ce contexte, plusieurs algorithmes ont été étudiés et mis au point. Ainsi, dans le cadre du positionnement de véhicule, la recherche s’est dirigée vers les méthodes robustes Hinfini et les méthodes par intervalles. Les méthodes Hinfini sont des méthodes linéaires prenant en compte une incertitude dans la modélisation et réalisant une optimisation min-max, c’est-à-dire minimisant une fonction de coût qui représente la pire situation compte tenu des incertitudes paramétriques. La contribution de ce travail concerne l’extension à des modèles faiblement non linéaires et l’utilisation d’une fenêtre glissante pour faire face à des mesures asynchrones. Les méthodes par intervalles développées ont pour but de calculer les couloirs de confiance des variables position et vitesse en se basant sur la combinaison d’intervalles issus des capteurs d’une part et sur l’utilisation conjointe d’un modèle dynamique et cinématique du véhicule d’autre part. Dans le cadre des capteurs logiciels pour bioprocédés, trois familles de méthodes ont été étudiées: le filtrage particulaire, les méthodes par intervalles et le filtrage par horizon glissant. Le filtrage particulaire est basé sur des méthodes de Monte-Carlo pour estimer la densité de probabilité conditionnelle de l’état connaissant les mesures. Un de ses principaux inconvénients est sa sensibilité aux erreurs paramétriques. La méthode développée s’applique aux bioprocédés et profite de la structure particulière des modèles pour proposer une version du filtrage particulaire robuste aux incertitudes des paramètres cinétiques. Des méthodes d’estimation par intervalles sont adaptées à la situation où les mesures sont disponibles à des instants discrets, avec une faible fréquence d’échantillonnage, en développant des prédicteurs appropriés. L’utilisation d’un faisceau de prédicteurs grâce à des transformations d’état et le couplage entre les prédicteurs avec des réinitialisations fréquentes permettent d’améliorer les résultats d’estimation. Enfin, une méthode basée sur le filtre à horizon glissant est étudiée en effectuant une optimisation min-max : la meilleure condition initiale est reconstruite pour le plus mauvais modèle. Des solutions sont aussi proposées pour minimiser la quantité de calculs. Pour conclure, les méthodes et résultats obtenus constituent un ensemble d’améliorations dans le cadre de la mise au point d’algorithmes robustes vis-à-vis des incertitudes. Selon les applications et les objectifs fixés, telle ou telle famille de méthodes sera privilégiée. Cependant, dans un souci de robustesse, il est souvent utile de fournir les estimations sous forme d’intervalles auxquels est associé un niveau de confiance lié aux conditions de l’estimation. C’est pourquoi, une des méthodes les plus adaptées aux objectifs de robustesse est représentée par les méthodes par intervalles de confiance et leur développement constituera un point de recherche futur. __________________________________________ Abstract : This thesis work is about the synthesis of robust state estimation methods applied to two different domains. The first area is dedicated to the safe positioning in transport. The objective is to compute the vehicle position and velocity by intervals with a great confidence level. The second area is devoted to the software sensor design in bioprocess applications. The component concentrations are estimated from a limited number of measurements and a mathematical model describing the dynamical behavior of the system. The main interest is to design algorithms which achieve estimation performance and take uncertainties into account coming from the model parameters and the measurement errors. In this context, several algorithms have been studied and designed. Concerning the vehicle positioning, the research activities have led to robust Hinfinity methods and interval estimation methods. The robust Hinfinity methods use a linear model taking model uncertainty into account and perform a min-max optimization, minimizing a cost function which describes the worst-case configuration. The contribution in this domain is an extension to some systems with a nonlinear model and the use of a receding time window facing with asynchronous data. The developed interval algorithms compute confidence intervals of the vehicle velocity and position. They use interval combinations by union and intersection operations obtained from sensors along with kinematic and dynamic models. In the context of bioprocesses, three families of state estimation methods have been investigated: particle filtering, interval methods and moving-horizon filtering. The particle filtering is based on Monte-Carlo drawings to estimate the posterior probability density function of the state variables knowing the measurements. A major drawback is its sensitivity to model uncertainties. The proposed algorithm is dedicated to bioprocess applications and takes advantage of the characteristic structure of the models to design an alternative version of the particle filter which is robust to uncertainties in the kinetic terms. Moreover, interval observers are designed in the context of bioprocesses. The objective is to extend the existing methods to discrete-time measurements by developing interval predictors. The use of a bundle of interval predictors thanks to state transformations and the use of the predictor coupling with reinitializations improve significantly the estimation performance. Finally, a moving-horizon filter is designed, based on a min-max optimization problem. The best initial conditions are generated from the model using the worst parameter configuration. Furthermore, additional solutions have been provided to reduce the computational cost. To conclude, the developed algorithms and related results can be seen as improvements in the design of estimation methods which are robust to uncertainties. According to the application and the objectives, a family may be favored. However, in order to satisfy some robustness criteria, an interval is preferred along with a measure of the confidence level describing the conditions of the estimation. That is why, the development of confidence interval observers represents an important topic in the future fields of investigation.

Nonlinear estimation techniques play an important role for process monitoring since some states and most of the parameters cannot be directly measured. There are many techniques available for nonlinear state and parameter estimation, i.e., extended Kalman filter (EKF), unscented Kalman filter (UKF), particle filtering (PF) and moving horizon estimation (MHE) etc. However, many issues related to the available techniques are to be solved. This dissertation discusses three important techniques in nonlinear estimation, which are the application of unscented Kalman filters, improvement of moving horizon estimation via computation of the arrival cost and different implementations of extended Kalman filters. First the use of several estimation algorithms such as linearized Kalman filter (LKF), extended Kalman filter (EKF), unscented Kalman filter (UKF) and moving horizon estimation (MHE) are investigated for nonlinear systems with special emphasis on UKF as it is a relatively new technique. Detailed case studies show that UKF has advantages over EKF for highly nonlinear unconstrained estimation problems while MHE performs better for systems with constraints. Moving horizon estimation alleviates the computational burden of solving a full information estimation problem by considering a finite horizon of the measurement data; however, it is non-trivial to determine the arrival cost. A commonly used approach for computing the arrival cost is to use a first order Taylor series approximation of the nonlinear model and then apply an extended Kalman filter. The second contribution of this dissertation is that an approach to compute the arrival cost for moving horizon estimation based on an unscented Kalman filter is proposed. It is found that such a moving horizon estimator performs better in some cases than if one based on an extended Kalman filter. It is a promising alternative for approximating the arrival cost for MHE. Many comparative studies, often based upon simulation results, between extended Kalman filters (EKF) and other estimation methodologies such as moving horizon estimation, unscented Kalman filter, or particle filtering have been published over the last few years. However, the results returned by the extended Kalman filter are affected by the algorithm used for its implementation and some implementations of EKF may lead to inaccurate results. In order to address this point, this dissertation investigates several different algorithms for implementing extended Kalman filters. Advantages and drawbacks of different EKF implementations are discussed in detail and illustrated in some comparative simulation studies. Continuously predicting covariance matrix for EKF results in an accurate implementation. Evaluating covariance matrix at discrete times can also be applied. Good performance can be expected if covariance matrix is obtained from integrating the continuous-time equation or if the sensitivity equation is used for computing the Jacobian matrix.

No description available.

Chemical vapor deposition (CVD) is an industrially important process to deposit crystalline and amorphous thin films on solid substrates. In situ sensing for CVD is necessary for process monitoring, fault detection, and process control. The challenge of in situ sensing lies in the prohibitive environment of the CVD process. Optical sensors such as the reflectometer and the ellipsometer are the most promising sensors because they can be installed outside of the deposition chamber, and are sensitive and easy to implement. However, the optical sensors do not measure film properties directly. Mathematical methods are needed to extract film properties from indirect optical measurements. Currently the most commonly used method is least squares fitting. In this project, we systematically investigated in situ reflectometry data interpretation based on state estimation theory. Optical models for light reflection on both smooth and rough surfaces were studied. The model validation results indicated that the effective medium model is better than the scalar scattering model when the surface is microscopically rough. The analysis of the observability for the sensor models indicated that the linearized observability does not always guarantee the true observability of a nonlinear system. We studied various state estimators such as batch least squares fitting (BLS), recursive least squares fitting (RLS), extended Kalman filter (EKF), and moving horizon estimation (MHE). It was shown that MHE is the general least-squares-based state estimation and BLS, RLS, and EKF are special cases of MHE. To reduce the computational requirement of MHE, a modified moving horizon estimator (mMHE) was developed which combines the advantage of the computational efficiency in RLS and the a priori estimate in MHE. State estimators were compared in simulated film growth processes, including both process model mismatch and sensor model mismatch, and reflection of both single wavelength and dual wavelength. In the case of process model mismatch and reflection on a smooth surface, there exists an optimum horizon size for both RLS and mMHE, although mMHE is less sensitive to the horizon size and performs better than RLS at all horizon sizes. The estimate with dual wavelength is more accurate than that with single wavelength indicating that estimation improves with more independent measurements. In the case of reflection on a rough surface, RLS failed to give a reasonable estimate due to the strong correlation between roughness and the extinction coefficient. However, mMHE successfully estimated the extinction coefficient and surface roughness by using the a priori estimate. MHE is much more computationally intensive than mMHE and there is no significant improvement on the estimation results. In the case of sensor model mismatch, either state estimator gave a good result, although mMHE consistently gave a better estimate, especially at a shorter horizon size. In order to test the state estimators in a real world environment, we built a cold-wall low-pressure chemical vapor deposition testbed with an in situ emissivity-correcting pyrometer. Fully automatic data-acquisition and instrument-control software was developed for the CVD testbed using LabVIEW. State estimators were compared using two experimental reflectance data sets acquired under different deposition conditions. The estimated film properties are compared with ex situ ellipsometry and AFM characterization results. In all cases, mMHE consistently yielded better estimates for processes under quite different deposition conditions. This indicated that mMHE is a useful and robust state estimator for in situ sensor data interpretation. By using the information from both the process and the sensor model, one can obtain a better estimate. A good feature of mMHE is that it provides such a versatile framework to organize all these useful information and gives a user the opportunity to interact with fitting and make wise decisions in the in situ sensor data interpretation.

This thesis considers the problem of control of nonlinear systems subject to limited availability of measurements and uncertainty in model parameters. To address this problem, first a linear offset free MPC is designed. Subsequently, a Lyapunov-based offset free MPC design is presented to handle structured uncertainty subject to constant disturbances. The controller's ability to handle unstructured uncertainty and measurement noise is demonstrated through simulation examples. Next, the problem of handling lack of state measurements as well as uncertainty is considered. To achieve simultaneous state and disturbance parameter estimation, a Lyapunov-based model predictive controller (MPC) is integrated with a moving horizon based mechanism, to achieve (where possible) offset elimination in the unmeasured states as well. A chemical reaction process example is presented to illustrate the key points. Finally its efficacy is demonstrated through a polymerization process example. / Thesis / Doctor of Philosophy (PhD)

The thesis work investigates brake friction estimation strategies. The friction between the brake disc and brake pads is not constant during the braking application and contributes to the amount of brake torque achieved at the wheels. In this study, it is considered that any change in the brake torque between the requested and achieved values is only due to the varying brake friction coefficient. The work gives three different approaches to estimate the brake friction coefficient using two prominent state estimation strategies, Unscented Kalman Filter and Moving Horizon Estimation. The inputs to the estimators are obtained from a Vehicle model, which is built using the wheel balance equations. The estimators have been tuned to minimize the estimation error in nominal conditions and tested for their robustness through a wide analysis, where the sensitivity of the strategies is checked against a spectra of potential system parameters and boundary conditions. Throughout all the analysis, the developed models estimate the brake friction coefficient within an acceptable error range. This work opens up opportunities for further studies that can be performed using the built estimator models. / Detta examensarbete studerar strategier för skattning av bromsfriktion. Friktionen mellan bromsskivan och bromsbeläggen är inte konstant under bromsförloppet och det är denna som genererar bromsmomentet för varje hjul. I detta arbete så antas att förändringen i bromsmoment mellan begärd och uppnått endast är på grund av varierande bromsfriktion mellan bromsbelägg och bromsskiva. Arbetet presenterar tre olika sätt att skatta bromsfriktionen genom användning av två kända skattningsmetoder, Uncented Kalman Filter och Moving Horizon Estimation. Ingående värden till skattningsmetoderna fås från en fordonsmodell som är byggd med hjälp av hjulbalansekvationer. Skattningsmetoderna har justerats så att de minimerar skattningsfelet i nominella fall och de är testade för robusthet genom en bred analys där känsligheten hos metoderna testas genom en flora av potentiella systemparametrar och gränsvärden. Genom hela analysen så uppnår de utvecklade skattningsmetoderna bromsfriktionsvärden med acceptabla felnivåer. Detta arbete öppnar upp för möjligheter för vidare analyser där de utvecklade metoderna kan användas.

L’objectif de ce travail a été de proposer une méthode de détection de défaut pour le déplacement longitudinal du chariot mobile du Radiotélescope de Nançay. L’importance de l’implémentation d’une procédure de détection des défauts a été mise en évidence grâce à la description des besoins du personnel en charge de la maintenance de cet instrument scientifique. Ce mémoire débute par un état de l’art sur différentes méthodes de diagnostic (détection et isolation des défauts), une analyse critique de ces méthodes est réalisée. Nous rappelons ensuite les notions d’observabilité avant de présenter l’observateur de Newton et l’observateur de Ciccarella. L’extension de ce dernier aux systèmes MIMO est réalisée par la suite. Une comparaison de ces différents observateurs termine le chapitre 2. Le chapitre 3 présente le Radiotélescope de Nançay et plus particulièrement la modélisation du déplacement longitudinal du chariot mobile. Une étude des propriétés du modèle est abordée par la suite. Le dernier chapitre traite de la validation partielle du modèle obtenu au chapitre précédent. Ensuite, l’utilisation des relations de redondances analytiques a permis de mettre en évidence la faisabilité du diagnostic sur l’application. L’utilisation de l’extension de l’observateur de Ciccarella pour le diagnostic permet de réaliser la détection de défaut capteur en utilisant un banc d’observateurs. Le suivi de paramètres du modèle permet de suivre l’évolution du système (vieillissement par exemple) et la détection de défaut actionneurs. Dans l’ensemble, les résultats obtenus permettent de conclure sur la bonne détection des défauts capteurs et actionneurs. / The aim of this work was to propose a fault detection method for the longitudinal displacement of the mobile cart of the Radiotelescope of Nançay. The importance of the implementation of a procedure of detection of the defects was highlighted thanks to the description the needs of the personnel in charge of maintenance for this scientific instrument. This memory begin with a state of the art on various methods of diagnosis (detection and isolation of the default), a critical analysis of these methods is carried out. We point out then the concepts of observability before presenting the Newton observer and the Ciccarella observer. The extension of this last to systems MIMO is carried out thereafter. A comparison of these various observers finishes chapter 2. Chapter 3 presents the Radiotelescope of Nançay and more particularly the modelling of the longitudinal displacement of the mobile cart. A study of the properties of the model is approached thereafter. The final chapter covers validation partial of the model obtained in the preceding Chapter. Then, the use of analytical redundancy relations made it possible to highlight the feasibility of the diagnosis on the application. The use of the extension of the Ciccarella observer for the diagnosis makes it possible to carry out the detection of sensor fault by using a bench of observers. The follow-up of parameters of the model makes it possible to follow the evolution of the system (ageing for example) and the detection of defect actuators. As a whole, the results obtained make it possible to conclude on good detection from the sensor and actuator faults.

Air combat is a complex situation, training for it and analysis of possible tactics are time consuming and expensive. In order to circumvent those problems, mathematical models of air combat can be used. This thesis presents air combat as a one-on-one influence diagram game where the influence diagram allows the dynamics of the aircraft, the preferences of the pilots and the uncertainty of decision making in a structural and transparent way to be taken into account. To obtain the players’ game optimal control sequence with respect to their preferences, the influence diagram has to be solved. This is done by truncating the diagram with a moving horizon technique and determining and implementing the optimal controls for a dynamic game which only lasts a few time steps. The result is a working air combat model, where a player estimates the probability that it resides in any of four possible states. The pilot’s preferences are modeled by utility functions, one for each possible state. In each time step, the players are maximizing the cumulative sum of the utilities for each state which each possible action gives. These are weighted with the corresponding probabilities. The model is demonstrated and evaluated in a few interesting aspects. The presented model offers a way of analyzing air combat tactics and maneuvering as well as a way of making autonomous decisions in for example air combat simulators.

The world's energy demand has been rapidly increasing and is projected to continue growing for at least the next two decades. With increasing global energy demand and competition from renewable energy, the oil and gas industry is striving for more efficient petroleum production. Many technical breakthroughs have enabled the drilling industry to expand the exploration to more difficult drilling such as deepwater drilling and multilateral directional drilling. For example, managed pressure drilling (MPD) offers ceaseless operation with multiple manipulated variables (MV) and wired drill pipe (WDP) provides two-way, high-speed measurements from bottom hole and along-string sensors. These technologies have maximum benefit when applied in an automation system or as a real-time advisory tool. The objective of this study is to investigate the benefit of nonlinear model-based control and estimation algorithms with various types of models. This work presents a new simplified flow model (SFM) for bottomhole pressure (BHP) regulation in MPD operations. The SFM is embedded into model-based control and estimation algorithms that use model predictive control (MPC) and moving horizon estimation (MHE), respectively. This work also presents a new Hammerstein-Wiener nonlinear model predictive controller for BHP regulation. Hammerstein-Wiener models employ input and output static nonlinear blocks before and after linear dynamics blocks to simplify the controller design. The control performance of the new Hammerstein-Wiener nonlinear controller is superior to conventional PID controllers in a variety of drilling scenarios. Conventional controllers show severe limitations in MPD because of the interconnected multivariable and nonlinear nature of drilling operations. BHP control performance is evaluated in scenarios such as drilling, pipe connection, kick attenuation, and mud density displacement and the efficacy of the SFM and Hammerstein-Wiener models is tested in various control schemes applicable to both WDP and mud pulse systems. Trusted high-fidelity drilling simulators are used to simulate well conditions and are used to evaluate the performance of the controllers using the SFM and Hammerstein-Wiener models. The comparison between non-WDP (semi-closed loop) and WDP (full-closed loop) applications validates the accuracy of the SFM under the set of conditions tested and confirms comparability with model-based control and estimation algorithms. The SFM MPC maintains the BHP within ± 1 bar of the setpoint for each investigated scenario, including for pipe connection and mud density displacement procedures that experience a wider operation range than normal drilling.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第24040号 / 情博第796号 / 新制||情||135(附属図書館) / 京都大学大学院情報学研究科システム科学専攻 / (主査)教授 加納 学, 教授 大塚 敏之, 教授 下平 英寿 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DGAM