The Compact Linear Collider (CLIC) is a proposal for a future linear electron--positron collider that could achieve collision energies of up to 3 TeV. In the CLIC concept the main high energy beam is accelerated using RF power extracted from a high intensity drive beam, achieving an accelerating gradient of 100 MV/m. This scheme places strict tolerances on the drive beam phase stability, which must be better than 0.2 degrees at 12 GHz. To achieve the required phase stability CLIC proposes a high bandwidth (>17.5 MHz), low latency drive beam "phase feedforward" (PFF) system. In this system electromagnetic kickers, powered by 500 kW amplifiers, are installed in a chicane and used to correct the phase by deflecting the beam on to longer or shorter trajectories. A prototype PFF system has been installed at the CLIC Test Facility, CTF3; the design, operation and commissioning of which is the focus of this work. Two kickers have been installed in the pre-existing chicane in the TL2 transfer line at CTF3 for the prototype. New optics have been created for the line to take these changes in to account, incorporating new constraints to obtain the desired phase shifting behaviour. Three new phase monitors have also been installed, one for the PFF input and two to verify the system performance. The resolution of these monitors must be significantly better than 0.2 degrees to achieve CLIC-level phase stability. A point by point resolution as low as 0.13 degrees has been achieved after a series of measurements and improvements to the phase monitor electronics. The performance of the PFF system depends on the correlation between the beam phase as measured at the input to the PFF system, and the downstream phase, measured after the correction chicane. Preliminary measurements found only 40% correlation. The source of the low correlation was determined to be energy dependent phase jitter, which has been mitigated after extensive efforts to measure, model and adjust the machine optics. A final correlation of 93% was achieved, improving the theoretical reduction in jitter using the PFF system from a factor 1.1 to a factor 2.7. The performance and commissioning of the kicker amplifiers and PFF controller are also discussed. Beam based measurements are used to determine the optimal correction timing. With a maximum output of around 650 V the amplifiers provide a correction range of ±5.5 ± 0.3 degrees. Finally, results from operation of the complete system are presented. A mean phase jitter of 0.28 ± 0.02 degrees is achieved, in agreement with the theoretical prediction of 0.27 ± 0.02 degrees for an optimal system with the given beam conditions. The current limitations of the PFF system, and possible future improvements to the setup, are also discussed.

Dans cette thèse, nous présentons des solutions pour la conception des systèmes de contrôle actif de vibrations. Dans la première partie, des méthodes de contrôle par action anticipatrice (feedforward) sont développées. Celles-ci sont dédiées à la suppression des perturbations bande large en utilisant une image de la perturbation mesurée par un deuxième capteur, en amont de la variable de performance à minimiser. Les algorithmes présentés dans cette mémoire sont conçus pour réaliser de bonnes performances et maintenir la stabilité du système en présence du couplage positif interne qui apparaît entre le signal de commande et l'image de la perturbation. Les principales contributions de cette partie sont l'assouplissement de la condition de « Stricte Positivité Réelle » (SPR) par l'utilisation des algorithmes d'adaptation « Intégrale + Proportionnelle » et le développement de compensateurs à action anticipatrice (feedforward) sur la base de la paramétrisation Youla-Kučera. La deuxième partie de la thèse concerne le rejet des perturbations bande étroite par contre-réaction adaptative (feedback). Une méthode d'adaptation indirecte est proposée pour le rejet de plusieurs perturbations bande étroite en utilisant des filtres Stop-bande et la paramétrisation Youla-Kučera. Cette méthode utilise des Filtres Adaptatifs à Encoche en cascade pour estimer les fréquences de perturbations sinusoïdales puis des Filtres Stop-bande pour introduire des atténuations aux fréquences estimées. Les algorithmes sont vérifiés et validés sur un dispositif expérimental disponible au sein du département Automatique du laboratoire GIPSA-Lab de Grenoble. / In this thesis, solutions for the design of robust Active Vibration Control (AVC) systems are presented. The thesis report is composed of two parts. In the first one, feedforward adaptive methods are developed. They are dedicated to the suppression of large band disturbances and use a measurement, correlated with the disturbance, obtained upstream from the performance variable by the use of a second transducer. The algorithms presented in this thesis are designed to achieve good performances and to maintain system stability in the presence of the internal feedback coupling which appears between the control signal and the image of the disturbance. The main contributions in this part are the relaxation of the Strictly Positive Real (SPR) condition appearing in the stability analysis of the algorithms by use of “Integral + Proportional” adaptation algorithms and the development of feedforward compensators for noise or vibration reduction based on the Youla-Kučera parameterization. The second part of this thesis is concerned with the negative feedback rejection of narrow band disturbances. An indirect adaptation method for the rejection of multiple narrow band disturbances using Band-Stop Filters (BSF) and the Youla-Kučera parameterization is presented. This method uses cascaded Adaptive Notch Filters (ANF) to estimate the frequencies of the disturbances' sinusoids and then, Band-stop Filters are used to shape the output sensitivity function independently, reducing the effect of each narrow band signal in the disturbance. The algorithms are verified and validated on an experimental setup available at the Control Systems Department of GIPSA-Lab, Grenoble, France.

Made available in DSpace on 2016-08-17T14:53:01Z (GMT). No. of bitstreams: 1 Fabio Nogueira da Silva.pdf: 1098466 bytes, checksum: a72dcced91748fe6c54f3cab86c19849 (MD5) Previous issue date: 2008-06-20 / FUNDAÇÃO DE AMPARO À PESQUISA E AO DESENVOLVIMENTO CIENTIFICO E TECNOLÓGICO DO MARANHÃO / We present in this work the results about two neural networks methods to solve the algebraic Riccati(ARE), what are used in many applications, mainly in the Linear Quadratic Regulator (LQR), H2 and H1 controls. First is showed the real symmetric form of the ARE and two methods based on neural computation. One feedforward neural network (FNN), that de¯nes an error as function of the ARE and a recurrent neural network (RNN), which converts a constrain optimization problem, restricted to the state space model, into an unconstrained convex optimization problem de¯ning an energy as function of the ARE and Cholesky factor. A proposal to chose the learning parameters of the RNN used to solve the ARE, by making a surface of the parameters variations, thus we can tune the neural network for a better performance. Computational experiments related with the plant matrices perturbations of the tested systems in order to perform an analysis of the behavior of the presented methodologies, that are based on homotopies methods, where we chose a good initial condition and compare the results to the Schur method. Two 6th order systems were used, a Doubly Fed Induction Generator(DFIG) and an aircraft plant. The results showed the RNN a good alternative compared with the FNN and Schur methods. / Apresenta-se nesta dissertação os resultados a respeito de dois métodos neuronais para a resolução da equação algébrica de Riccati(EAR), que tem varias aplicações, sendo principalmente usada pelos Regulador Linear Quadrático(LQR), controle H2 e controle H1. É apresentado a EAR real e simétrica e dois métodos baseados em uma rede neuronal direta (RND) que tem a função de erro associada a EAR e uma rede neuronal recorrente (RNR) que converte um problema de otimização restrita ao modelo de espaço de estados em outro de otimização convexa em função da EAR e do fator de Cholesky de modo a usufruir das propriedades de convexidade e condições de otimalidade. Uma proposta para a escolha dos parâmetros da RNR usada para solucionar a EAR por meio da geração de superfícies com a variação paramétrica da RNR, podendo assim melhor sintonizar a rede neuronal para um melhor desempenho. Experimentos computacionais relacionados a perturbações nos sistemas foram realizados para analisar o comportamento das metodologias apresentadas, tendo como base o princípio dos métodos homotópicos, com uma boa condição inicial, a partir de uma ponto de operação estável e comparamos os resultados com o método de Schur. Foram usadas as plantas de dois sistemas: uma representando a dinâmica de uma aeronave e outra de um motor de indução eólico duplamente alimentado(DFIG), ambos sistemas de 6a ordem. Os resultados mostram que a RNR é uma boa alternativa se comparado com a RND e com o método de Schur.

Les actionneurs piézoélectriques font partie des outils les plus utilisés dans les applications à l'échelle micro/nano-métrique (micromanipulation, microassemblage, micropositionnement, etc…). Du point de vue fonctionnel, on distingue les actionneurs mono-axe (permettant d'obtenir la déflection suivant une direction) et les actionneurs multi-axes (pouvant fléchir suivant plusieurs directions). La notoriété des actionneurs piézoélectriques est due à un certain nombre de performances telles qu'une large bande passante (plus du kHz possible), une très bonne résolution (de l'ordre du nanomètre), une faible consommation en énergie électrique, une grande densité de force, une facilité d'alimentation et d'intégration, etc. Cependant, ces actionneurs sont caractérisés par des non-linéarités fortes (hystérésis et la dérive lente), des oscillations mal-amorties, et sont sensibles à la variation des conditions ambiantes (en particulier à la variation de la température). Pour les actionneurs multi-axes, il s'ajoute un problème des couplages entre les différents axes de l'actionneur. Cette thèse propose des stratégies innovantes de commande des actionneurs piézoélectriques multi-axes pour contrer les problèmes sus-mentionnés. Ces stratégies sont groupées en deux catégories. La première catégorie concerne les techniques de commande en boucle fermée. Ces techniques sont les plus adaptées pour garantir la robustesse et un niveau de précision élevé pour les actionneurs piézoélectriques. Cependant, à l'échelle micro/nano-métrique, ces techniques sont limitées par un manque d'espace suffisant pour installer des capteurs de position. La deuxième catégorie concerne la commande en boucle ouverte dont l'intérêt majeur est lié au fait qu'il n'y a pas besoin de capteurs pour la commande, ce qui constitue un avantage en terme de coût et facilité d'intégration. Dans cette thèse, nous proposons d'abord les techniques de modélisation et de commande en boucle ouverte multivariables. Ensuite, nous faisons une analyse des effets de la température sur les actionneurs piézoélectriques et nous proposons des techniques de commande en boucle ouverte et en boucle fermée de ces effets. Enfin, une stratégie de commande en boucle fermée par découplage, visant à obtenir des correcteurs d'ordre réduit pour les actionneurs multi-axes est proposée. Toutes ces techniques sont vérifiées et appliquées expérimentalement à un actionneur piézoélectrique de type tube. / Piezoelectric actuators are among the most used tools in many applications at micro/nano-scale (micromanipulation,microassembly, micropositioning, etc). From a functional perspective, there exist mono-axis actuators(which are made to bend in one direction) and multi-axis actuators (which provide deflections in different directions).The popularity of piezoelectric actuators is especially due to their high resolution (nanometric resolution),the large bandwidth (greater than 1kHz possible), the low electrical power consumption, the high force density,the ease of integration in positioning systems, etc. However, piezoelectric actuators are characterized by hysteresisand creep nonlinearities, badly damped vibrations and they are sensitive to the variation of ambient conditions(especially to the temperature variation). In addition, multi-axis actuators exhibit cross-couplings betweentheir axis. This thesis proposes novel strategies for modeling and control of multi-axis piezoelectric actuators,with the aim to counteract the aforementionned problems. These strategies are grouped into two categories.The first category concerns feedback control techniques. These techniques are the most suitable to ensurethe robustness and a high level of precision for piezoelectric actuators. However, at the micro/nanoscale, thesetechniques are limited by the lack of enough physical space to install feedback sensors. The second categoryconcerns the feedforward control techniques. The main advantage of these techniques is related to the factthat, in feedforward control schemes, feedback sensors are not needed for tracking. This allows to achieve ahigh degree of packageability and the cost reduction. In this thesis, we first propose multivariable modelingand feedforward control techniques. Then, we analyse the effects of temperature variation on piezoelectricactuators and we propose feedforward and feedback control techniques for these effects. Finally, a feedbackstrategy based on decoupling techniques with an aim to reduce the order of feedback controllers for multi-axispiezoelectric actuators, is proposed. All these modeling and control strategies are experimentally applied on apiezoelectric tube actuator.

This thesis describes the process of rebuilding an experimental model of a single pendulum on a cart into the double pendulum on a cart. The control algorithm in MATLAB/Simulink environment for stabilization of the pendulum in the inverse position is designed. For this purpose, LQR state feedback control was implemented. Also method for swinging the pendulum into inverse position from stable state (swing-up) was designed. Feedforward method was utilised for swing-up control. In the thesis, functionality of these algorithms is shown.

The subject of this thesis is neural network acceleration with the goal of reducing the number of floating point multiplications. The theoretical part of the thesis surveys current trends and methods used in the field of neural network acceleration. However, the focus is on the binarization techniques which allow replacing multiplications with logical operators. The theoretical base is put into practice in two ways. First is the GPU implementation of crucial binary operators in the Tensorflow framework with a performance benchmark. Second is an application of these operators in simple image classifier. Results are certainly encouraging. Implemented operators achieve speed-up by a factor of 2.5 when compared to highly optimized cuBLAS operators. The last chapter compares accuracies achieved by binarized models and their full-precision counterparts on various architectures.

Neuronové sítě vycházející z architektury WaveNet a sítě využívající rekurentní vrstvy jsou v současnosti používány jak pro syntézu lidské řeči, tak pro „black box“ modelování systémů pro úpravu akustického signálu – modulační efekty, nelineární zkreslovače apod. Úkolem studenta bude shrnout dosavadní poznatky o možnostech využití neuronových sítí při modelování akustických signálů. Student dále implementuje některý z modelů neuronových sítí v programovacím jazyce Python a využije jej pro natrénování a následnou simulaci libovolného efektu nebo systému pro úpravu akustického signálu. V rámci semestrální práce vypracujte teoretickou část práce, vytvořte zvukovou databázi pro trénování neuronové sítě a implementujte jednu ze struktur sítí pro modelování zvukového signálu. Neuronové sítě jsou v průběhu posledních let používány stále více, a to víceméně přes celé spektrum vědních oborů. Neuronové sítě založené na architektuře WaveNet a sítě využívající rekurentních vrstev se v současné době používají v celé řadě využití, zahrnující například syntézu lidské řeči, nebo napřklad při metodě "black-box" modelování akustických systémů, které upravují zvukový signál (modulačí efekty, nelineární zkreslovače, apod.). Tato akademická práce si dává za cíl poskytnout úvod do problematiky neuronových sítí, vysvětlit základní pojmy a mechanismy této problematiky. Popsat využití neuronových sítí v modelování akustických systémů a využít těchto poznatků k implementaci neuronových sítí za cílem modelování libovolného efektu nebo zařízení pro úpravu zvukového signálu.

Die Arbeit befasst sich mit der Entwicklung automatischer Lenkungen, die die von Schienenfahrzeugen bekannte Spurtreue auf n-gliedrige, mehrachsgelenkte Straßenfahrzeuge übertragen. Spurtreu bedeutet folglich, dass die Lenkachsmittelpunkte keinen seitlichen Versatz zueinander aufweisen. Dafür wird ein modellbasiertes automatisches Lenkverfahren systematisch konzipiert, entworfen und erprobt, das sowohl eine vollautomatische Spurführung als auch eine halbautomatische Nachführung erlaubt. Die modellbasierten automatischen Lenkungen unterliegen keinen praktisch relevanten Einschränkungen. Das wird durch die Verwendung einer Steuerungsstruktur mit zwei Freiheitsgraden erreicht, die aus einer modellbasierten Vorsteuerung und einem Rückführregler besteht. In der Vorsteuerung werden die Lenkwinkel aller Achsen berechnet, mit denen der Sollweg theoretisch spurtreu befahren wird. Durch den Einsatz eines speziell angepassten, modularen Mehrkörpermodells gelingt diese Berechnung allgemein für eine Klasse n-gliedriger Fahrzeuge. Zum Ausgleich von nicht vermeidbaren Modellunbestimmtheiten und nicht gemessenen Störungen werden ein nichtlinearer Mehrgrößenregler sowie achs-individuelle lineare Eingrößenregler entworfen und miteinander verglichen. Simulationen und Fahrversuche zeigen, dass das entwickelte Verfahren in einem weiten Geschwindigkeitsbereich robust gegenüber typischen Einflussgrößen wie Fahrbahn- und Beladungszustand ist.

The development of machine learning research has provided statistical innovations and further developments within the field of time series analysis. This study seeks to investigate two different approaches on artificial neural network models based on different learning techniques, and answering how well the neural network approach compares with a basic autoregressive approach, as well as how the artificial neural network models compare to each other. The models were compared and analyzed in regards to the univariate forecast accuracy on 20 randomly drawn time series from two different time frequencies from the M4 competition dataset. Forecasting was made dependent on one time lag (t-1) and forecasted three and six steps ahead respectively. The artificial neural network models outperformed the baseline Autoregressive model, showing notably lower mean average percentage error overall. The Multilayered perceptron models performed better than the Long short-term memory model overall, whereas the Long short-term memory model showed improvement on longer prediction time dimensions. As the training were done univariately  on a limited set of time steps, it is believed that the one layered-approach gave a good enough approximation on the data, whereas the added layer couldn’t fully utilize its strengths of processing power. Likewise, the Long short-term memory model couldn’t fully demonstrate the advantagements of recurrent learning. Using the same dataset, further studies could be made with another approach to data processing. Implementing an unsupervised approach of clustering the data before analysis, the same models could be tested with multivariate analysis on models trained on multiple time series simultaneously.

We consider a mechanistic non-linear machine learning approach to learning signals in financial time series data. A modularised and decoupled algorithm framework is established and is proven on daily sampled closing time-series data for JSE equity markets. The input patterns are based on input data vectors of data windows preprocessed into a sequence of daily, weekly and monthly or quarterly sampled feature measurement changes (log feature fluctuations). The data processing is split into a batch processed step where features are learnt using a Stacked AutoEncoder (SAE) via unsupervised learning, and then both batch and online supervised learning are carried out on Feedforward Neural Networks (FNNs) using these features. The FNN output is a point prediction of measured time-series feature fluctuations (log differenced data) in the future (ex-post). Weight initializations for these networks are implemented with restricted Boltzmann machine pretraining, and variance based initializations. The validity of the FNN backtest results are shown under a rigorous assessment of backtest overfitting using both Combinatorially Symmetrical Cross Validation and Probabilistic and Deflated Sharpe Ratios. Results are further used to develop a view on the phenomenology of financial markets and the value of complex historical data under unstable dynamics.