• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 3
  • Tagged with
  • 3
  • 3
  • 3
  • 2
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Automated modeling and implementation of power converters on a real-time FPGA-based emulator

De Cuyper, Kevin 07 December 2015 (has links) (PDF)
Designing a new power electronic conversion system is a multi-step process that requires the R\&D team(s) to go through an extended prototyping phase whose goal is to validate the design in its nominal state, as well as to test its behavior when it is subjected to abnormal conditions. To properly and safely validate all devices that are external to the power stage itself, such as the controllers and the protection systems, one of the best-suited device is a real-time emulator of the converter circuit, a platform that obeys the same mathematical laws and produces the same signals as the original device withoutactually realizing the power conversion. Unfortunately, these models are often based on analog solvers which are difficult to build, must be redesigned for each modification and are subject to drift and aging. While multiple digital real-time emulators have appeared on the market in the last decades, they typically require powerful and expensive computing platforms to perform their calculations or are not generic enough to emulate the more complex power circuits. In this work, we present a new framework that allows the rapid prototyping of a wide range of power converters by translating a power converter schematic drawn on a computer to a real-time equivalent set of equations which is processed by an FPGA with an emulation time-step of less than one microsecond. Contrary to the previously published works, our tools enable the use of entry-level FPGAs even for the emulation circuits composed of twenty switches or more. This framework takes the form of a tool-chain that starts by extracting the necessary information and a standard description from the initial circuit. However, due to the intricate ways in which the switches and diodes can change their state, this raw information is too complex to be processed and emulated directly.Our first major contribution to the state of the art is a way to automatically analyze these changes in order to reduce the complexity of the problem as much as possible while keeping all the necessary information intact. In this thesis, we develop two tools that are able to find all possible changes in the state of the switches that may appear in the immediate future, thereby reducing the quantity of information required to emulate the circuit. Thanks to the global optimization provided by our tools, simulating a typical AC-to-DC converter composed of 12 switches could require 80\% less resources when compared to existing emulators.To enable the emulation or large power converters, we have created a partitioning method which divides the circuit in multiple sub-circuits which are analyzed and optimized separately. The performances of this partitioning are demonstrated by the emulation of a three-phase three-level converter with a relative error of a less that 5% on the signals.To handle our new framework, a dedicated digital platform has been developed. In order to provide the best results even on small FPGAs, particular attention is given to the low resources usage and the low latency of our design. Through multiple examples, we show that this inexpensive real-time emulation platform is able to accurately emulate many circuits in open- or closed-loop operation with a sampling rate higher than 1 MHz / Doctorat en Sciences de l'ingénieur et technologie / info:eu-repo/semantics/nonPublished
2

Study of electrical interfaces for electrostatic vibration energy harvesting / Étude d'interfaces électriques pour les récupérateurs d'énergie vibratoire électrostatiques

Karami, Armine 16 May 2018 (has links)
Les récupérateurs d'énergie vibratoire électrostatiques (REV) sont des systèmes convertissant une partie de l'énergie cinétique de leur environnement en énergie électrique, afin d'alimenter de petits systèmes électroniques. Les REV inertiels sont constituées d'un sous-système mécanique bâti autour d'une masse mobile, ainsi que d'une interface électrique. Ces deux blocs sont couplés par un transducteur électrostatique. Cette thèse étudie l'amélioration des performances des REV par la conception optimisée de leur interface électrique. La première partie de cette thèse étudie une famille d'interfaces électriques appelées pompes de charge (PC). On commence par la construction d'une théorie formelle des PC. Des interfaces rapportées dans la littérature sont identifiées comme membres de cette famille. Cette dernière est ensuite complétée par une nouvelle topologie de PC. Une comparaison des différents PC est alors faite dans le domaine électrique, puis un outil semi-analytique est présenté pour la comparaison des PC en prenant en compte le couplage électromécanique. L'étude des PC se termine par la présentation d'une nouvelle méthode de mesure du potentiel d'électret des REV. La deuxième partie de la thèse présente une approche de conception radicalement différente de ce qui est présenté dans les travaux actuels sur les REV. Elle préconise une synthèse active de la dynamique de la masse des REV à travers leur interface électrique. Nous montrons d'abord que cela permet la conversion d'énergie en quantités proches des limites physiques, et ce à partir de vibrations d'entrée de forme arbitraire. Enfin, une architecture pour un tel REV est proposée et testée en simulation. / Electrostatic vibration energy harvesters (e-VEHs) are systems that convert part of their surroundings' kinetic energy into electrical energy, in order to supply small-scale electronic systems. Inertial E-VEHs are comprised of a mechanical subsystem that revolves around a mobile mass, and of an electrical interface. The mechanical and electrical parts are coupled by an electrostatic transducer. This thesis is focused on improving the performances of e-VEHs by the design of their electrical interface. The first part of this thesis consists in the study of a family of electrical interfaces called charge-pumps conditioning circuits (CPCC). It starts by building a formal theory of CPCCs. State-of-the-art reported conditioning circuits are shown to belong to this family. This family is then completed by a new CPCC topology. An electrical domain comparison of different CPCCs is then reported. Next, a semi-analytical tool allowing for the comparison of CPCC-based e-VEHs accounting for electromechanical effects is reported. The first part of the thesis ends by presenting a novel method for the measurement of e-VEHs' built-in electret potential. The second part of the thesis presents a radically different design approach than what is followed in most of state-of-the-art works on e-VEHs. It advocates for e-VEHs that actively synthesize the dynamics of their mobile mass through their electrical interface. We first show that this enables to convert energy in amounts approaching the physical limits, and from arbitrary types of input vibrations. Then, a complete architecture such an e-VEH is proposed and tested in simulations submitted to human body vibrations.
3

A deep learning theory for neural networks grounded in physics

Scellier, Benjamin 12 1900 (has links)
Au cours de la dernière décennie, l'apprentissage profond est devenu une composante majeure de l'intelligence artificielle, ayant mené à une série d'avancées capitales dans une variété de domaines. L'un des piliers de l'apprentissage profond est l'optimisation de fonction de coût par l'algorithme du gradient stochastique (SGD). Traditionnellement en apprentissage profond, les réseaux de neurones sont des fonctions mathématiques différentiables, et les gradients requis pour l'algorithme SGD sont calculés par rétropropagation. Cependant, les architectures informatiques sur lesquelles ces réseaux de neurones sont implémentés et entraînés souffrent d’inefficacités en vitesse et en énergie, dues à la séparation de la mémoire et des calculs dans ces architectures. Pour résoudre ces problèmes, le neuromorphique vise à implementer les réseaux de neurones dans des architectures qui fusionnent mémoire et calculs, imitant plus fidèlement le cerveau. Dans cette thèse, nous soutenons que pour construire efficacement des réseaux de neurones dans des architectures neuromorphiques, il est nécessaire de repenser les algorithmes pour les implémenter et les entraîner. Nous présentons un cadre mathématique alternative, compatible lui aussi avec l’algorithme SGD, qui permet de concevoir des réseaux de neurones dans des substrats qui exploitent mieux les lois de la physique. Notre cadre mathématique s'applique à une très large classe de modèles, à savoir les systèmes dont l'état ou la dynamique sont décrits par des équations variationnelles. La procédure pour calculer les gradients de la fonction de coût dans de tels systèmes (qui dans de nombreux cas pratiques ne nécessite que de l'information locale pour chaque paramètre) est appelée “equilibrium propagation” (EqProp). Comme beaucoup de systèmes en physique et en ingénierie peuvent être décrits par des principes variationnels, notre cadre mathématique peut potentiellement s'appliquer à une grande variété de systèmes physiques, dont les applications vont au delà du neuromorphique et touchent divers champs d'ingénierie. / In the last decade, deep learning has become a major component of artificial intelligence, leading to a series of breakthroughs across a wide variety of domains. The workhorse of deep learning is the optimization of loss functions by stochastic gradient descent (SGD). Traditionally in deep learning, neural networks are differentiable mathematical functions, and the loss gradients required for SGD are computed with the backpropagation algorithm. However, the computer architectures on which these neural networks are implemented and trained suffer from speed and energy inefficiency issues, due to the separation of memory and processing in these architectures. To solve these problems, the field of neuromorphic computing aims at implementing neural networks on hardware architectures that merge memory and processing, just like brains do. In this thesis, we argue that building large, fast and efficient neural networks on neuromorphic architectures also requires rethinking the algorithms to implement and train them. We present an alternative mathematical framework, also compatible with SGD, which offers the possibility to design neural networks in substrates that directly exploit the laws of physics. Our framework applies to a very broad class of models, namely those whose state or dynamics are described by variational equations. This includes physical systems whose equilibrium state minimizes an energy function, and physical systems whose trajectory minimizes an action functional (principle of least action). We present a simple procedure to compute the loss gradients in such systems, called equilibrium propagation (EqProp), which requires solely locally available information for each trainable parameter. Since many models in physics and engineering can be described by variational principles, our framework has the potential to be applied to a broad variety of physical systems, whose applications extend to various fields of engineering, beyond neuromorphic computing.

Page generated in 0.0447 seconds