Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Decentralized Current Control for an Open-Winding Synchronous Machine with Local Estimation of the Electromotive Force
No full textInternational audienceThis article proposes a decentralized current control law for open winding multiphase permanent magnet synchronous machines. An active disturbance rejection control scheme based on an extended state observer is proposed. The proposed observer enhances the current control performance of the flatness-based control already proposed in the literature, mainly caused by the inverse model limitation. This new decentralized control structure was evaluated in a wide range of operating points and compared to the flatness-based control law in simulation to validate its robustness
Etat de l’art sur les méthodes de détection de changement et application pour le suivi de chute de blocs
No full textInternational audienceThe acceleration of global warming is increasing the frequency of major gravitational events, making it increasingly necessary to monitor their precursors (minor events). While in situ sensors (interferometers, seismometers, etc.) are reliable, their large-scale deployment remains limited by logistical and economic factors. Terrestrial time-lapse cameras, which are low-cost and easily interpretable, offer an interesting alternative. However, their automatic processing remains challenging due to drastic changes in appearance and lighting conditions.Recent advances in computer vision and artificial intelligence now make it possible to envision robust automation for the detection of serac and rock falls. In this presentation, we will review the state of the art of various change detection methods available in the literature, as applied to gravitational events. Given the low frequency of these hazards and the difficulty of annotation (requiring expertise and time-consuming analysis), particular attention will be paid to weakly supervised and/or zero-shot learning, which are more realistic in our application context.While the latest change detection methods based on foundation models are impressive for satellite data and autonomous driving, they show serious limitations with our data. In contrast, deep matching methods appear to be a more relevant option. This study highlights the current challenges in rockfall detection and outlines research perspectives that could be pursued to address the problem.L’accélération du réchauffement climatique intensifie l’occurrence d’événements gravitaires majeurs, rendant la surveillance de leurs précurseurs (événements mineurs) de plus en plus nécessaire. Si les capteurs in situ (interféromètres, sismomètres, etc.) se montrent fiables, leur déploiement à grande échelle reste limité par des facteurs logistiques et économiques. Les caméras time-lapse terrestres, peu coûteuses et facilement interprétables, constituent une alternative intéressante. Cependant, leur exploitation automatique reste délicate, en raison des changements d’aspect et de luminosité drastiques pouvant survenir. Les récentes avancées en vision par ordinateur et en intelligence artificielle permettent néanmoins d’envisager l’automatisation robuste de la détection des chutes de séracs et de blocs. Dans cet exposé, nous réaliserons un état de l’art des différentes méthodes disponibles dans la littérature pour la détection de changements, appliquées aux événements gravitaires. Etant donné la faible fréquence de ces aléas et la difficulté d’annotation (expertise nécessaire, analyse chronophage), une attention particulière sera portée à l’apprentissage faiblement supervisé et/ou ” zero-shot ”, plus réaliste dans notre contexte applicatif. Les dernières méthodes en détection de changement basées sur les modèles de fondation sont impressionnantes sur la donnée satellitaire et pour la conduite autonome, mais montrent de sérieuses limites sur nos données. Au contraire, les méthodes de mise en correspondance profondes semblent être une option plus pertinente. Cette étude nous permet de mettre en évidence les défis actuels dans le domaine de la détection de chutes de blocs et des perspectives de recherche qui pourront être envisagées pour répondre au problème
III-V Heterostructure Plasma Etching for Low Cost and High Efficiency Photonic Devices
No full textInternational audienc
Excavators Electrohydraulic Systems Steady-State Optimal Flow Metering Computation Using a Multiple-Integer Non-Linear Programming (MINLP) Solver
No full textInternational audienceAbstract This paper introduces a new approach to compare hydraulic distribution architectutes by computing the steady-state optimal flow metering of electrohydraulic power transmission systems for multi-actuated machine applications such as excavators. The method is based on a constrained optimization problem formulation with binary variables used to handle non-linearites through piecewise linearization. The proposed formulation allows for the study of optimal control and energy efficiency performances of a large variety of electrohydraulic systems. As an exemple of the proposed method, an excavator electrohydraulic system is modelled before computing its optimal metering over the actuators speed and load trajectories from a measured Volvo excavator digging duty-cycle. Then simulated results are inspected, notably showing an energy efficiency of 36.1 % in regards with the Volvo excavator load-sensing hydraulic distribution efficiency which is 24.3 %
Étude de la fiabilité des GaN SP-HEMT sous surtensions en commutation dure
No full textInternational audienc
Dimensionnement optimal d'une architecture d'électronique de puissance pour le chauffage par induction impulsionnel. Application au contrôle non destructif par thermographie infrarouge
No full textPoster demandé (doctorant en début de 1ère année)International audienceLe procédé de chauffage par induction couplé à la thermographie infrarouge est un moyen de détecter des défauts de manière non destructive dans l’industrie de production des matériaux métalliques. Cependant à l’heure actuelle, les systèmes utilisés pour la chauffe par induction dans le cadre de la thermographie active n’ont pas été conçus pour cette utilisation et ne répondent donc pas correctement aux spécificités de celle-ci. La performance de ce contrôle est étroitement liée au régime transitoire lié à l’électronique de puissance utilisée dans les générateurs par induction, ce qui fait l’objet de ces travaux. Ce papier propose donc d’étudier trois topologies de chauffe par induction dans le cadre de la thermographie pour connaître les spécificités de chacune d’un point de vue électronique de puissance
Image processing and deep learning for the characterization of polluted soils and their petrophysical properties
No full textInternational audienceIn France, over 350000 sites are potentially polluted, including 60000 in Auvergne-Rhône-Alpes. Most of these sites are contaminated by organic liquid contaminants, such as hydrocarbons and PCBs (Polychlorobiphenyls). The current method of pollution diagnosis, based on soil sampling, is costly and time-consuming. Moreover, this method frequently leads to errors in the assessment of pollutants and volumes of contaminated soil, impacting on the choice of remediation techniques, the cost and duration of treatments. To this end, GINGER R&D team aims to develop an in-situ image processing technique for estimating pollution and soil petrophysical properties, providing near-real-time quantitative data. In this context, a classification model based on embeddings generated by neural networks is being developed, making it possible to exploit multiple descriptors such as color, and those derived from neural embeddings. These descriptors will be computed on a variety of modalities, including color images and from different acquisition sources. This deep learning model, developed from a database created during the thesis, will be evaluated on several levels: controlled-condition images, cubitainer data and data from real sites. In this work, we present our initial results obtained by fine-tuning a pre-trained model to classify images into lithological classes and water saturation levels, enhancing polluted site characterization and remediation efforts
Impact of Shift-Angle Topologies on the Performance of PCB-Embedded Solenoid Inductors
No full textInternational audienceThe integration of magnetic components within printed circuit boards (PCBs) offers an innovative pathway to enhance electromagnetic performance while reducing size and improving manufacturability in power electronics. This study investigates the impact of shift-angle topologies in PCB-embedded solenoid inductors, focusing on minimizing electromagnetic losses to enhance high-frequency performance for applications such as compact DC/DC converters and power modules. Four angular configurations were analyzed, assessing how different trace alignments influence flux distribution, eddy current generation, and inductance behavior. To isolate the effects of angle shift, the study employed Finite Element Method (FEM) simulations under direct current (DC) excitation, eliminating the influence of skin effect, proximity effect, core loss, and parasitic capacitance. Results showed that modifying the shift angle significantly altered flux density-especially the perpendicular (Z) component responsible for eddy current formation. Among the DC-tested topologies, Configurations with two counter wise angles in the coils top-layer and bottom-layer traces demonstrated the lowest flux loss and minimal eddy current dissipation. Experimental validation confirmed the simulation findings, with electromagnetic losses within a 6% margin. Tests under alternating current (AC) conditions using an impedance analyzer verified the benefits of reverse angle configurations but revealed that the angle position affects the resonance frequency
Determining Relevant 3D Roughness Parameters for Sandblasted Surfaces: A Methodological Approach
No full textInternational audienceThis study presents a robust methodology for analyzing 3D roughness parameters to characterize sandblasted surfaces, identifying the most relevant descriptors for process optimization. Sandblasting with irregularly shaped corundum particles is performed using five grit sizes (25, 50, 90, 125, and 250 µm) and three pressure levels (2, 3, and 4 bar). The resulting surfaces are characterized through eight 3D roughness parameters: Sa, Spc, Sal, Sfd, Sdq, Sdr, Spd, and Str. A linear model of the form Q = a + b.D + d.D.P, where Q represents the roughness parameter, D is the average grit size, and P is the sandblasting pressure, is employed. For Spd, a nonlinear model, Spd = (a + b.D + d.D.P)2, yields a significantly improved determination coefficient, demonstrating the model’s enhanced ability to capture the complexity of the Spd parameter. The double-bootstrap analysis validates the statistical significance of all models, providing confidence intervals for each parameter. This approach emphasizes the importance of advanced 3D roughness descriptors for accurately analyzing surface textures in sandblasting processes, offering a reliable framework for surface characterization and industrial optimization
