Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Empreinte Énergétique de l'Apprentissage Distribué dans l'IoT : une Approche Conjointe Basée sur l'Émulation Matériel-Logiciel
No full textNational audienceL'Intelligence Artificielle Distribuée dans le Continuum Informatique rencontre des défis liés aux exigences computationnelles et de communication de l'apprentissage sur les dispositifs en périphérie. Pour y remédier, nous présentons FLIoT, un émulateur d'apprentissage distribué basé sur ns-3, intégrant des interactions réseau réelles et une évaluation énergétique fondée sur des mesures matérielles empiriques de l'ESP32-C6 DKC-1. FLIoT, testé sur des données de bâtiments intelligents avec le framework Flower, facilite l'évaluation de la consommation énergétique en permettant la création de signatures énergétiques applicatives, accélérant ainsi le prototypage dans les environnements d'apprentissage distribué
Using SiC JBS diode as voltage clamping in SSCB application: performances and limitations
No full textInternational audienceThe performances and limitation of SiC JBSdiodes used as clamping devices for SSCB havebeen investigated trough simulation andexperimental measurements. The simulationsevaluations are based on CAD and electro-thermal circuit modelling of packaged devices.The simulations have been done on a wideblocking voltage range, form 80V to 1.7kV. Anoptimal trade-off in terms of layout and avalanchecapability has been found. A failure criterion interms of power density has been established,linked to device top metal layers melting (around500°C) and measured on commercial devices.Avalanche to Nominal current Ratio variesbetween 6 up to 10 depending on themanufacturer. A LTspice electro-thermal modelallows to predict the required SiC devices area toachieve specific performances. Finally, ademonstrator of 1.2kV-2x50A-SOT227 SiC JBSclamping diode is presented
A temporal dependency preserving approach for anomaly detection on multivariate time series
No full textInternational audienc
Multi-line input network for class EF inverter with enhanced design flexibility
No full textInternational audienceThis paper addresses a new input network for the class EF inverter, which utilizes multiple transmission lines instead of the commonly used quarter-wave line. Experimental demonstrations at 15 MHz with 40 W output power are conducted, showcasing that enhanced performance can be achieved with the proposed input network
Mechanical models and simulations of PEMFCs: A comprehensive review
No full textInternational audienceIn recent years, there has been considerable progress in developing models concerning the electrochemical, thermal, and mass transport processes of the Proton Exchange Membrane Fuel Cell (PEMFC). This advancement has resulted in numerous models being documented in the literature, alongside several review articles that aim to categorise them. However, this extensive body of literature does not have a clear definition or a dedicated review to help identify and classify mechanical models of PEMFCs. This article first focuses on identifying key characteristics that define mechanical models for PEMFCs. It then provides a comprehensive classification of these models for each component and more complex structures. To achieve this, an original representation highlights essential model information and traces their chronological evolution. This approach clearly articulates what has been accomplished and what still needs to be addressed, serving as a valuable resource for PEMFC simulation and future research
NetBone Meets LLMs: Toward Explainable Backbone Extraction
No full textInternational audienceDespite recent progress in standardizing backbone extraction through tools like NetBone, interpreting the results remains challenging. This is especially true for non-expert users unfamiliar with the method assumptions. In this work, we extend NetBone to support interpretation through Large Language Models (LLMs). To support this, we propose a metadata-driven pipeline that automatically captures each step of the backbone extraction process, including network descriptions, method parameters, filtering settings, and structural properties. This structured metadata is used to prompt an LLM to generate contextual and method-aware explanations. We believe this extension would improve interpretability and helps users understand how method choices affect the structure and meaning of their results
Bias in Federated Learning: A Comprehensive Survey
No full textInternational audienceFederated Learning (FL) enables collaborative model training over multiple clients’ data, without sharing these data for better privacy. Addressing bias in FL remains a challenge. In this paper, we irst present a taxonomy of FL bias, presenting the causes and the diferent types of FL bias, namely demographic bias, performance-related bias, and contribution-related bias. We then categorize FL bias mitigation, in terms of used methods and provided guarantees, before providing a comprehensive and comparative analysis of existing works. Finally, we highlight key challenges and open research directions, including the impact of FL bias mitigation on model utility, privacy, and robustness
Empirical model for the dynamic pressures exerted by a tunnel boring machine: numerical application to a rocky greenfield environment
No full textInternational audienc
Étude comparative des méthodologies de commande MIMO pour la réduction des vibrations dans des systèmes multi-piézoélectriques à auto-détection
No full textThis work addresses the challenge of vibration control in a flexible plate using two piezoelectric patches in a Self-Sensing Actuator (SSA) configuration, meaning they simultaneously serve as sensors and actuators. Two H ∞ MIMO control strategies, centralized and structured, are developed and compared, accounting for the coupling effects between the piezoelectric elements. Additionally, a third strategy based on a previously developed SISO methodology is also included for comparative analysis. The identification-based modeling approach coupled with model reduction techniques enable the design of a low-order controller, achieving significant vibration reduction within a specified frequency range.Ce travail aborde le défi du contrôle des vibrations dans une plaque flexible en utilisant deux patchs piézoélectriques dans une configuration d'Actionneur Auto-Sensible (SSA), ce qui signifie qu'ils servent simultanément de capteurs et d'actionneurs. Deux stratégies de commande H ∞ MIMO, centralisée et structurée, sont développées et comparées, en tenant compte des effets de couplage entre les éléments piézoélectriques. De plus, une troisième stratégie basée sur une méthodologie SISO précédemment développée est également incluse pour l'analyse comparative. L'approche de modélisation basée sur l'identification, couplée à des techniques de réduction de modèle, permet la conception d'un contrôleur de faible ordre, permettant d'obtenir une réduction significative des vibrations dans une plage de fréquences spécifiée
Electrical Characterization of Modular 3D Packaging Assembled with Compressed Metal Foams
No full textInternational audienceThis paper proposes a novel modular packaging concept based on the 3D integration of 1.2kV SiC MOSFET with metal foams for electrical contacts with pressure assembly. These foams could allow improving the disassembly of switching cells by keeping low electrical resistance and inductance inside the package. Thus, static and dynamic experimental characterizations are carried out to evaluate their impact compared to a massive interconnection represented by copper interposers. Static measurements show that the electrical resistance of the package is 2.3mΩ which represents 7% of the total resistance. Dynamic characterizations demonstrate that the stray inductance of the switching cell is equal to 2.7nH. With results of the massive interconnection close to the ones of foams, the material used for the electrical interconnection have no impact on the electrical performances of the power module. The modification of pressure assembly of the power module shows however a global increase of the packaging resistance and stray inductance of the power module for both solutions. This increase is more important for the massive interconnection due the non-deformable mechanicals properties