Buffer tanks are widely used within the process industry to prevent flow variations from being directly propagated throughout a plant. The capacity of the tank is used to smoothly transfer inlet flow upsets to the outlet. Ideally, the tank thus works as a low pass filter where the available tank capacity limits the achievable flow smoothing. For infrequently occurring upsets, where the system has time to reach steady state between flow changes, the averaging level control problem has been extensively studied. After an inlet flow change, flow filtering has traditionally been obtained by letting the tank level deviate from its nominal value while slowly adapting the outlet to cancel out the flow imbalance and eventually bringing back the level to its set-point. The system is then again in steady state and ready to surge the next upset. By ensuring that the single largest upset can be handled without violating the level constraints, satisfactory flow smoothing is obtained. In this thesis, the smoothing of frequently changing inlet flows is addressed. In this case, standard level controllers struggle to obtain acceptable flow smoothing since the system rarely is in steady state and flow upsets can thus not be treated as separate events. To obtain a control law that achieves optimal filtering while directly accounting for future upsets, the averaging level control problem was approached using robust model predictive control (MPC). The robust MPC differs in the way it obtains flow smoothing by not returning the tank level to a fixed set-point. Instead, it lets the steady state tank level depend on the current value of the inlet flow. This insight was then used to propose a linear control structure, designed to filter frequent upsets optimally. Analyses and simulation results indicate that the proposed linear and robust MPC controller obtain flow smoothing comparable to the standard optimal averaging level controllers for infrequent upsets while handling frequent upsets considerably better.

Dans cette thèse, nous nous intéressons au problème de détection de mesures GPS erronées. En effet, en zones urbaines, les acquisitions sont fortement dégradées par des phénomènes de multi-trajets ou de multiples réflexions des signaux avant d’arriver à l’antenne réceptrice. En forêt, de multiples obstacles bloquent les signaux satellites, ce qui diminue la redondance des mesures. Alors que les algorithmes présents dans les récepteurs GPS détectent au maximum une mesure erronée par pas de temps, avec une combinaison de différents systèmes de navigation, l’hypothèse d’une seule erreur à la fois n’est plus tenable et la détection et gestion des données erronées (défaillantes, aberrantes ou outliers selon les différentes terminologies) représente un enjeu majeur dans les applications de navigation autonome et de localisation robuste et devient un nouveau défi technologique.La contribution principale de cette thèse est un algorithme de détection de mesures de pseudo-distances aberrantes exploitant la modélisation a contrario. Deux critères fondés sur l’espérance du nombre de fausses alarmes (NFA) sont utilisés pour mesurer la cohérence d’un ensemble de mesures sous l’hypothèse d’un modèle de bruit.Notre seconde contribution concerne l’introduction des mesures Doppler dans le processus de localisation. Nous étendons la détection d’outliers conjointement dans les mesures de pseudo-distance aux mesures Doppler et proposons une localisation par couplage avec le filtre particulaire soit SIR soit de Rao-Blackwell qui permet d’estimer analytiquement la vitesse.Notre troisième contribution est une approche crédibiliste pour la détection des mesures aberrantes dans les pseudo-distances. S’inspirant du RANSAC, nous choisissons, parmi les combinaisons d’observations possibles, la plus compatible selon une mesure de cohérence ou d’incohérence. Une étape de filtrage évidentiel permet de tenir compte de la solution précédente. Les approches proposées donnent de meilleures performances que les méthodes usuelles et démontrent l’intérêt de retirer les données aberrantes du processus de localisation. / In this work, we focus on the problem of detection of erroneous GPS measurements. Indeed, in urban areas, acquisitions are highly degraded by multipath phenomena or signal multiple reflections before reaching the receiver antenna. In forest areas, the satellite occlusion reduces the measurements redundancy. While the algorithms embedded in GPS receivers detect at most one erroneous measurement per epoch, the hypothesis of a single error at a time is no longer realistic when we combine data from different navigation systems. The detection and management of erroneous data (faulty, aberrant or outliers depending on the different terminologies) has become a major issue in the autonomous navigation applications and robust localization and raises a new technological challenge.The main contribution of this work is an outlier detection algorithm for GNSS localization with an a contrario modeling. Two criteria based on number of false alarms (NFA) are used to measure the consistency of a set of measurements under the noise model assumption.Our second contribution is the introduction of Doppler measurements in the localization process. We extend the outlier detection to both pseudo-ranges and Doppler measurements, and we propose a coupling with either the particle filter SIR or the Rao-Blackwellized particle filter that allows us to estimate analytically the velocity.Our third contribution is an evidential approach for the detection of outliers in the pseudo-ranges. Inspired by the RANSAC, we choose among possible combinations of observations, the most compatible one according to a measure of consistency or inconsistency. An evidential filtering step is performed that takes into account the previous solution. The proposed approaches achieve better performance than standard methods and demonstrate the interest of removing the outliers from the localization process.

Intensity modulated proton therapy (IMPT) in head and neck squamous cell carcinoma (HNSCC) offers superior advantages over conventional photon therapy, by generating high conformal doses to the target volume and improved sparing of the organ at risks (OARs). Besides, robust treatment planning approaches, which account for uncertainties directly into the plan optimization process, are able to generate high quality plans robust against uncertainties compared to a PTV margin expansion approach. During radiation treatment, patients are prone to present anatomical variations during the treatment course, which can be random deviations in patient positioning, as well as treatment-induced tumor shrinkage and patient weight variations. For IMPT plans using a PTV margin expansion, these anatomical variations might disturb the calculated nominal plan, with a decrease to the dose delivered to the target volume and/or increased dose to the OARs above its tolerance, and a plan adaptation might be needed. However, the influence of these anatomical variations in robustly optimized plans for HNSCC entities has not been determined. The first part of this thesis compared two proton therapy methods, single-field optimization (SFO) and multi-field optimization (MFO), applied to the treatment of unilateral HNSCC target volumes, consisting of a cohort of 8 patients. For each method, a PTV-based and a robustly optimized plan were generated, resulting in four plans per patient. The four plans showed adequate target coverage on the nominal plan, with larger doses to the ipsilateral parotid gland for both SFO approaches. No plan showed a clear advantage when variations in the anatomy during the treatment course were considered, and the same was observe considering additional setup and range uncertainties. Hence, no plan showed a decisive superiority regarding plan robustness and potential need of replanning. In the second part of this thesis, an anatomical robustly optimized plan approach was proposed (aRO), which considers additional CT datasets in the plan optimization, representing random non-rigid patient positioning variations. The aRO approach was compared to a classical robustly optimized plan (cRO) and a PTV-based approach for a cohort of 20 bilateral HNSCC patients. PTV-based and cRO approaches were not sufficient to account for weekly anatomical variations, showing a degradation in the target coverage in 10 and 5 of 20 cases, respectively. Conversely, the proposed aRO approach was able to preserve the target coverage in 19 of 20 cases, with only one patient requiring plan adaptation. An extended robustness analysis conducted on both cRO and aRO plan approaches considering weekly anatomical variations, setup and range errors, showed that the variations in anatomy were the most critical variable for loss in target coverage, while setup and range uncertainties played a minor role. The price of the increased plan robustness for the aRO approach was a significant larger integral dose to the healthy tissue, compared to the cRO plan. However, the increase in integral dose was not reflected on the planned dose to the OARs, which were comparable between both plans. Therefore, the price for a superior plan robustness can be considered as low. In the current clinical practice, the implementation of the aRO approach would be able to reduce the need of plan adaptation. For its application, the acquisition of additional planning CT datasets, considering a complete patient repositioning between scans is required, in order to simulate random non-rigid position variations as simulated in this study by the use of the first two weekly cCTs in the plan optimization. Further studies using multiple planning CT acquisition, including strategies to reduce the patient CT dose such as dual-energy CT and iterative reconstruction algorithms, are needed to confirm the presented findings. Additionally, the aRO approach applied to other body sites and entities might also be investigated. In near future, further in-room imaging methods such as cone-beam CT and magnetic resonance imaging, optimized for proton therapy, might be used to acquire additional datasets. Moreover, alternative approaches capable of modeling variations in patient positioning as biomechanical models and deep learning methods might be able to generate in silico additional image datasets for use in proton treatment planning. In summary, this thesis proposes an additional contribution for robust treatment planning in IMPT, with the generation of treatment plans robust against anatomy variations, together with setup and range uncertainties, which can benefit the clinical workflow by reducing the need of plan adaptation.:Contents List of Figures List of Tables List of Abbreviations 1 Introduction 2 Proton Therapy 2.1 Rationale for Proton Therapy 2.2 Beam Delivery Techniques 2.2.1 Passive Scattering 2.2.2 Pencil Beam Scanning 2.3 Uncertainties in Proton Therapy 2.3.1 Target Volume Definition 2.3.2 Range Uncertainty 2.3.3 Setup Uncertainty 2.3.4 Biological Uncertainty 2.3.5 Anatomical Variations 3 Robust Treatment Planning and Robustness Evaluation 3.1 Robust Treatment Planning 3.1.1 Including Uncertainties in the Optimization 3.1.2 Differences Between Approaches 3.2 Robustness Evaluation 3.2.1 Error Scenarios 3.2.2 Visual Evaluation of Plan Robustness 3.2.3 Summary 4 Illustration of Robust Treatment Planning in a Simple Geometry 4.1 Plan Design 4.2 Plan Results 4.2.1 Doses on Nominal Plan 4.2.2 Influence of Uncertainties in Plan Robustness 4.3 Discussion and Conclusion 5 Evaluation of Robust Treatment Plans in Unilateral Head and Neck Squamous Cell Carcinoma 5.1 Study Design 5.1.1 Calculation Parameters 5.1.2 Plan Robustness Evaluation 5.2 Results 5.2.1 Evaluation of Nominal Plan Doses 5.2.2 Evaluation of Plan Robustness Against Uncertainties 5.3 Discussion 5.4 Conclusions 6 Assessment of Anatomical Robustly Optimized Plans in Bilateral Head and Neck Squamous Cell Carcinoma 6.1 Anatomical Robust Optimization 6.2 Study Design 6.2.1 Calculation Parameters 6.2.2 Assessment of Plan Robustness 6.3 Results 6.3.1 Evaluation of Nominal Plan Doses 6.3.2 Evaluation of Plan Robustness Against Uncertainties 6.4 Discussion 6.4.1 Robustness Against Anatomical Variations 6.4.2 Robustness Against Additional Setup and Range Uncertainties 6.4.3 Study Limitations 6.5 Conclusions 7 Summary 8 Zusammenfassung Bibliography Appendix

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第23545号 / 情博第775号 / 新制||情||132(附属図書館) / 京都大学大学院情報学研究科数理工学専攻 / (主査)教授 太田 快人, 教授 山下 信雄, 教授 大塚 敏之 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

The aim of this work is to design and create a robust speech activity detector that is able to detect speech in different languages, in a noise environment and with music on background. I decided to solve this problem by using a neural network as a classification model that assigns one of the four possible classes - silence, speech, music, or noise to the input of audio recording. The resulting tool is able to detect the speech in at least 12 languages. Speech with musical background up to 88 % accuracy and system success on noisy data reaches from 84 % (5 dB SNR) to 88 % (20 dB SNR). This tool can be used for speech activity detection in various research areas of speech processing. The main contribution is the elimination of music, which when not eliminated, significantly increases the error rate of systems for speaker identification or speech recognition.

The doctoral thesis deals with mobile laser scanning and data processing. Mobile laser scanning was carried out in the Moravian Karst in Suchy zleb and Pusty zleb at various seasons. Further scanning was carried out at the Training Forest Enterprise Masaryk Forest Křtiny. The Riegl VMX-450 scanner was used to collect 3D data. The data files were processed in the OPALS software. For the extraction of ground and non-ground points from the laser cloud, a combination of hierarchical interpolation and robust filtering was used. Subsequently, the terrain points were used to model digital terrain models with grids 0.05 x 0.05 m, 0.25 x 0.25 m, 1 x 1 m. Furthermore, digital models from Moravian Karst were compared with the Digital relief model of the Czech Republic 4G and reference points. Digital models from Masaryk Forest Křtiny were compared with photogrammetric and terrestrial data. The high accuracy of the derived DTMs is about 0.10 m. The high quality of the derived DTM can be used to monitor and analyze field changes and morphological structures.

This dissertation embodies a comprehensive approach towards resilient monitoring of frid events using Structure Preserving Energy Functions (SPEFs) and introduces a novel control architecture in Multi Terminal Direct Current (MTDC) grids, for inter-area oscillation damping and achieving robustness to AC as well as DC side, post-contingency events in the modern power grid. This work is presented as a collection of several publications which investigate and address the proposed research topics. At first, SPEFs are derived for multi-machine IEEE benchmark models with the help of the Wide-Area Measurements (WAMs). A physics-based hybrid approach to develop one-to-one mapping between properties of energy function components with respect to the type of fault in the system is introduced. The proposed method is tested offline on a IEEE-39 bus, New England Test System (NETS), with particular interest in monitoring the most sensitive energy functions during relay misoperations. Such events can be precipitated by zone 3 trips in distance relays due to load encroachment during stressed conditions. These might include a genuine misoperation, a false trip due to cyber-attacks, or a load encroachment, all of which are undesirable under normal operating circumstances. An online monitoring scheme is introduced in an actual blackout simulation in the Western Electricity Coordinating Council (WECC) to examine what further indications these energy function components can provide, especially during a cascading sequence, and how they could supervise critical tripping decisions by distance relays. Next, a futuristic grid comprised of Voltage Source Converter (VSC) based AC-MTDC is considered due to its recent proliferation in integrating offshore wind farms to onshore grids, and additionally improving strength of weak AC grids. A robust control is designed using the converter station poles as actuators to provide damping support to the surrounding AC grid. Further, a design problem is envisioned and implemented that introduces disturbance rejection into control architecture by designing a novel explicitly modeled disturbance plant in the Linear Matrix Inequality (LMI) framework. Finally, a novel robust inter-area oscillation damping controller is designed that proves its effectiveness in inter-area mode settling times, and provides robustness to (n-1) contingencies in the AC as well as the DC side of the meshed AC-MTDC grid.

In Finance, the use of Automated Trading Systems (ATS) on markets is growing every year and the trades generated by an algorithm now account for most of orders that arrive at stock exchanges (Kissell, 2020). Historically, these systems were based on advanced statistical methods and signal processing designed to extract trading signals from financial data. The recent success of Machine Learning has attracted the interest of the financial community. Reinforcement Learning is a subcategory of machine learning and has been broadly applied by investors and researchers in building trading systems (Kissell, 2020). In this thesis, we address the issue that deep reinforcement learning may be susceptible to sampling errors and over-fitting and propose a robust deep reinforcement learning method that integrates techniques from reinforcement learning and robust optimization. We back-test and compare the performance of the developed algorithm, Robust DDPG, with UBAH (Uniform Buy and Hold) benchmark and other RL algorithms and show that the robust algorithm of this research can reduce the downside risk of an investment strategy significantly and can ensure a safer path for the investor’s portfolio value.

Les actionneurs piézoélectriques sont très utilisés pour les systèmes de positionnement pour des tâches à l'échelle micro/nanométrique en raison de leur haute résolution (sub-micrométrique), leur grande bande passante, et une densité de force élevée. Cependant, ils sont caractérisés par des non-linéarités telles que l'hystérésis et la dérive lente, par une grande sensibilité à l’environnement, et pour certains par des comportements oscillatoires. Ces caractéristiques ont un impact considérable sur les tâches que ces actionneurs doivent effectuer et leur contrôle reste souvent difficile. Différents correcteurs robustes ont été développés pour contrôler les actionneurs piézoélectriques. Il s'agit notamment de l'approche grand gain, des approches H-inf, des approches de contrôle basée sur l'observation des perturbations, ...etc. Ces techniques ont démontré une amélioration significative des performances mais mènent souvent à des correcteurs d'ordre élevé qui sont difficiles à mettre en œuvre. Cette thèse consiste à développer des correcteurs pour des actionneurs piézoélectriques en combinant l'analyse d'intervalle et les techniques classiques de commande.Les avantages principaux d’utiliser des intervalles est qu'ils permettent de modéliser facilement les incertitudes paramétriques en les limitant par des bornes. Par ailleurs, les travaux existants démontrent qu’il est possible de synthétiser de manière plus simplifiée des correcteurs robustes d’ordre faible, c-à-d ordre inférieur à celui du modèle. L'état de l'art sur l'utilisation des techniques par intervalle pour la synthèse de correcteurs peut être présenté en deux catégories : les techniques par intervalle basées sur des fonctions de transfert, et les techniques par intervalle basées sur la représentation d’état. Les techniques par intervalle basées sur les fonctions de transfert sont actuellement limitées pour modéliser et synthétiser des correcteurs pour des systèmes monovariables incertains tandis que les techniques basées sur la représentation d'état sont bien adaptées pour des systèmes multivariables incertains. Néanmoins, ces travaux existants pour des systèmes multivariables sont limités aux modèles avec des matrices d'état et d'entrée de structures spéciales. En outre, elles ne portent que sur le degré de stabilité du système en boucle fermée et ne prennent pas donc en compte des spécifications sur les performances. Cette thèse développe des outils de synthèse de correcteurs robustes pour des systèmes multivariables à incertitudes paramétriques dans l’approche d’état par intervalle sans structure particulière et en considérant à priori des performances. Des validations expérimentales sur différents actionneurs piézoélectriques, et ce en commande en position et en force, sont efféctuées. Enfin, la thèse propose également l’extension des observateurs à entrée inconnue pour les systèmes par intervalle afin de compléter les techniques de commande proposées. / Piezoelectric actuators are widely used at micro/nanoscale because of their simpleconfiguration, high resolution (sub-nanometric), high speed (large bandwidth upto 1kHz), and high force density. However, they are characterized by some nonlinearitiessuch as hysteresis, internal friction and creep,...etc. These characteristicsconsiderably impact the dynamics of the piezoactuators which makes the controlof these systems not a trivial task. Various robust controllers have been developedto control piezoelectric actuators. These include high gain feedback approach, H1approach, disturbance observer based control approach,...etc. Those techniquesdemonstrated a significant improvement of the control performance, but they oftenderive controllers with high-order which are difficult for implementation. Tobypass this limitation, we focus on the thesis on combining interval analysis withclassical controller design techniques to obtain a low order controllers. The mainadvantage of intervals is that they permit to model parametric uncertainties easilyby bounding them. Furthermore, the process of modeling the system uncertaintiesby intervals makes the synthesis of robust controller with low order relatively easy.The state of the art on the use of interval techniques to design and derive robustcontrollers for uncertain system can be divided into two categories: intervaltransfer functions based approaches and interval state-space representation basedapproaches. Interval transfer functions based designs have been widely used tomodel and to control SISO (Single Input single Output) systems subjected to uncertainties.These approaches make the synthesis of robust controllers for suchsystems easy with providing good performance. However, the current work thatuse interval transfer functions are limited to systems in SISO case. In the otherside, the state-space based approaches have been shown to be well adapted tosynthesis robust controllers for multivariable systems. Nevertheless, the excitingworks are limited to systems with state and input matrices of special structures.Furthermore, they address only the degree of stability of the closed-loop systemwithout discussing performance specification. In order to make the design of robustcontroller using interval state-space approach possible for any interval state-spacestructure, this thesis will explore the interval state-space control design using robustpole assignment technique. This proposed approach will guarantee the stability and the desired performance of the closed-loop system also it allows to obtaina low order controller.For this matter, an algorithm based on Set Inversion Via Interval Analysis(SIVIA) combined with interval eigenvalues computation is proposed to seek for aset of robust gains. This recursive SIVIA-based algorithm allows to approximatewith subpaving the set solutions [K] that satisfy the inclusion of the eigenvaluesof the closed-loop system in a desired region in the complex plan. Furthermore,simple algorithms are proposed to find the optimal feedback gains among the rangeof robust gains [K] as well as the range of the gains that satisfy input constraints,all with the help of interval analysis. Finally, in order to improve the controllerperformance, we were directed our attention to nonlinear control approaches andespecially interval sliding mode control (ISMC) design using interval observers.The effectiveness of the proposed approaches are tested by a real experimentationon several platforms developed in our laboratory to achieve robust performance.

Ce travail s'inscrit dans le cadre d'une thèse CIFRE - Convention Industrielle de Formation - supervisée par la laboratoire LAMIH UMR CNRS 8201 Laboratoire et financé par l'entreprise SRDEP (Société de Ressources et de Développement pour les Entreprises et les Particuliers) et l'ANRT (Association Nationale de la Recherche et de la Technologie).Il s'agit d'une thèse applicative dans le domaine de la commande robuste numérique des systèmes biomécaniques. L'objectif est de concevoir et de réaliser un appareil de ré-entraînement à l'effort physique pour contrôler les paramètres physiologiques de l'homme avec plusieurs finalités : diagnostic, l'optimisation des prescriptions et sécurisation de l'exercice.Le travail réalisé de divise en deux parties :1 - L'étude scientifique : Dans cette partie, un modèle non linéaire de fréquence cardiaque est proposé ainsi qu'une méthodologie d'identification. A partir des résultats obtenus dans l'étape précédente, est développé un algorithme de commande robuste. Ce dernier est validé à partir de plusieurs tests réalisés en temps réel sur l'homme à l'aide d'un prototype.2 - Réalisation et industrialisation : Cela comprend la conception de l'appareil, le dimensionnement des différentes parties, la conception du système embarqué temps réel, la fabrication des cartes électroniques, l'implantation des algorithmes de commandes ainsi que la gestion des relations avec les partenaires et les sous-traitants.Il s'agit donc clairement d'un projet multidisciplinaire incluant de la mécanique, de l'électronique, de l'informatique, de l'automatique mais aussi le design, des relations clients-fournisseurs, des démarches de contacts avec des clients potentiels et de nombreux essais et mises au point. Dans cette thèse, seuls les aspects scientifiques seront discutés, en particulier ceux centrés autour des techniques de l'automatique : Modélisation, identification, commande et diagnostic même si l'ensemble des activités non liées à ces aspects ont pris un certain temps. / This work of PhD Thesis is realized in the framework of "CIFRE" scholarship supervised by LAMIH UMR CNRS 8201 laboratory and financed by SRDEP (Société de Ressources et de Développement pour les Entreprises et les Particuliers) Company and the ANRT (Association Nationale de la Recherche et de la Technologie) organization. It falls in the domain of numerical identification and robust control of biomedical systems. The aim is to design and to realize a prototype of physical training apparatus based on an electrically controlled brake to regulate some physiological parameters (namely heart rate) by means of exercise intensity. This might be used for diagnostic, optimization of exercise prescriptions and in making exercise more secure for patients.The work can be divided into two parts:1 - Scientific study : in this part, a nonlinear model is proposed for the heart rate/exercise intensity system as well as an identification process. Then, a robust control strategy is established and tested in real time on several subjects using thedeveloped prototype.2 - Realization of embedded system : This includes the design of electromechanical system; the real time embedded system, electronic cards as well as the implementation of control algorithms.It is clearly a multidisciplinary project including mechanics, electronics, informatics, control systems, physiology of human exercise besides project management, relations with partners, and direction of team work. In this thesis only scientific aspects will be discussed and especially those relevant to system identification and control, despite the fact that a relatively big percentage of time had to be devoted to the other aspects.