HAL Portal Univ-Côtedazur
Not a member yet
95391 research outputs found
Sort by
Comment devenir un insider sur un nouveau marché à l’international ?
No full textWebinaire organisé par le Think Tank « La Fabrique de l’Exportation » : https://www.youtube.com/watch?v=XKmpCXQRBysCe webinaire porte sur le développement d'une ETI (entreprise de taille intérmédiaire) allemande sur le marché français. Les intervenants expliquent comment l'entreprise est parvenue à surmonter les difficultés pour devenir un insider local
Approches robustes et informées par la physique pour la modélisation et l'estimation de la dynamique véhicule sur des horizons longs
No full textSimulating and identifying nonlinear dynamical systems is a central task in many engineering applications, including robotics, transportation, and control. In the context of vehicle dynamics, accurate long-horizon prediction is essential for performance evaluation, safety analysis, and system design. However, these systems often involve strong nonlinearities, complex interactions, and limited or noisy measurement data, which makes both model calibration and state estimation challenging. This thesis explores hybrid modeling approaches that combine physics-based models with data-driven learning to address these limitations. The first part of the work investigates multi-step loss functions for learning dynamical models from data. These losses encourage models to capture long-term system behavior and improve robustness, especially in systems with dissipative characteristics. A bias-variance analysis is provided to explain the improved generalization performance, particularly in contractive settings, which are common in physical systems such as vehicles. In the second part, a physics-informed learning framework is developed. Starting from a known simulation model of a vehicle, a neural network is pretrained on synthetic data and fine-tuned on real on-road measurements. This transfer learning approach improves generalization in scenarios where data is sparse or noisy. Although the method is demonstrated on lateral vehicle dynamics, it remains applicable to other physical systems with partially known dynamics. The third contribution addresses the problem of state estimation, where the goal is to reconstruct latent states from partial and noisy observations. A new method is proposed based on deep learning and Moving Horizon Estimation (MHE). Applied to vehicle simulation setting, the model learns an inverse mapping offline and enables real-time state estimation without requiring explicit optimization or access to system equations at runtime. Overall, the contributions presented in this thesis support the development of robust modeling techniques tailored to complex dynamical systems, with demonstrated benefits in vehicle simulation.La modélisation et l'identification de systèmes dynamiques non linéaires constituent un enjeu central dans de nombreux domaines de l'ingénierie, tels que la robotique, les transports ou les systèmes de contrôle. Dans le cadre des dynamiques véhicule, la capacité à effectuer des prédictions précises à long terme est essentielle pour l'évaluation des performances, l’analyse de la sécurité et la conception des systèmes. Ces systèmes présentent souvent de fortes non- linéarités, des interactions complexes entre composants, ainsi que des mesures limitées ou bruitées, rendant la calibration des modèles et l'estimation d'état particulièrement difficiles. Cette thèse explore des approches hybrides combinant modèles physiques et apprentissage automatique afin de surmonter ces limitations. La première partie du travail s'intéresse à l'utilisation de fonctions de coût multi-pas pour l’apprentissage de modèles dynamiques à partir de données. Ces pertes favorisent la capture du comportement à long terme des systèmes et améliorent la robustesse, notamment dans les systèmes présentant des caractéristiques dissipatives. Une analyse biais-variance est proposée pour expliquer les gains observés en termes de généralisation, en particulier dans les systèmes contractants, couramment rencontrés dans les applications physiques telles que les véhicules. La deuxième partie développe un cadre d'apprentissage informé par la physique. En partant d’un modèle de simulation connu d'un véhicule, un réseau de neurones est pré-entraîné sur des données synthétiques, puis ajusté à partir de mesures issues de situations réelles. Cette approche par transfert d'apprentissage améliore la capacité de généralisation dans des contextes où les données sont rares ou bruitées. Bien que la méthode soit appliquée à la dynamique latérale de véhicules, elle reste généralisable à d'autres systèmes physiques partiellement connus. La troisième contribution porte sur l'estimation d'état, dont l'objectif est de reconstruire les états latents d'un système à partir d'observations partielles et bruitées. Une nouvelle méthode est proposée, fondée sur l'apprentissage profond et inspirée de l'estimation par horizon glissant (Moving Horizon Estimation, MHE). Appliqué à un contexte de simulation véhicule, le modèle apprend hors ligne une application inverse, permettant ainsi une estimation d'état en temps réel, sans avoir besoin d'optimisation explicite ni d'accès aux équations dynamiques lors de l’inférence. Les contributions présentées dans cette thèse participent au développement de méthodes de modélisation robustes adaptées aux systèmes dynamiques complexes, avec des bénéfices démontrés dans le domaine de la simulation véhicule
III-nitride semiconductors are more flexible than other compound semiconductors for integrated photonics and efficient nonlinear interactions: Invited Paper
No full textInternational audienceIII-nitride semiconductors like GaN and AlN have very specific properties that can be useful for integrated photonic circuits. We discuss some recent results associated with single-arm Mach-Zehnder interferometers, sign reversal of second-order nonlinear susceptibility and integration of two-dimensional materials
High-Order Asymptotic-Preserving IMEX schemes for an ES-BGK model for Gas Mixtures
No full textIn this work we construct a high-order Asymptotic-Preserving (AP) Implicit-Explicit (IMEX) scheme for the ES-BGK model for gas mixtures introduced in [16]. The time discretization is based on the IMEX strategy proposed in [26] for the single-species BGK model and is here extended to the multi-species ES-BGK setting. The resulting method is fully explicit, uniformly stable with respect to the Knudsen number and, in the fluid regime, it reduces to a consistent and high-order accurate solver for the limiting macroscopic equations of the mixture. The IMEX structure removes the stiffness associated with the relaxation term so that the time step is constrained only by a hyperbolic CFL condition. The full solver couples a high-order space and velocity discretization that includes third-order time integration, a CWENO3 finite-volume reconstruction in space, exact conservation of macroscopic moments in the discrete velocity space, and a multithreaded implementation. The proposed approach can handle an arbitrary number of species. Its accuracy and robustness are demonstrated on a set of multidimensional kinetic tests for gas mixtures, where the AP property and the correct asymptotics are numerically verified across different regimes
Development of a flight simulator for the WEST plasma and control system
No full textInternational audienceThe ITER project (www.iter.org) should demonstrate in the next decades the technical feasibility of controlled fusion reactions in tokamaks. One of the critical issues reaching this purpose is the design of plasma scenarios and associated controllers in order to achieve the desired performance while satisfying the operational limits. To succeed, the non-linearity, the uncertainties, and the limited observability of the plasma presently require adjusting controllers and scenarios during commissioning sessions. This method is time-consuming and must be reduced to the strict minimum time. To address this issue, the community has developed for several years simulation tools to design both controllers and scenarios using numerical models of the plasma. From simple linear models of the vertical plasma instability to integrated modeling of both plasma transport and equilibrium, these codes are now efficient enough to predict the plasma behavior and be called “flight simulator”. In this article, the flight simulator developed for WEST will be presented. One of the main features is the use as input of the same pulse schedule files and the same controllers as in the WEST Plasma Control System (PCS). Based on the free boundary equilibrium code NICE with flux diffusion equation and a 1D transport model, a consistent plasma time evolution can be computed and reduces the risk of failure due to numerical issues. To illustrate the abilities of the tool, a standard WEST X-point formation will be simulated and compared to the real data
Young learner autonomy in synchronous oral telecollaborative tasks: Participants, arena, and turns
No full textInternational audienceThis study examines the potential of synchronous oral telecollaborative tasks to foster learner autonomy among young language learners. Previous research highlights both the language learning affordances of technology-mediated task-based language teaching and challenges for successful implementation with young learners and suggests that learner autonomy is an important mediating variable. The present article investigates autonomy by exploring learner participation during task-as-process and the teacher’s role in creating opportunities for learning in technology-mediated exchanges. We propose a new analytical framework based on the notion of arena , drawing on Goffman’s dramaturgical concept of frontstage versus backstage interaction, to inform a fine-grained investigation of turn-taking during the same task-as-workplan implemented in two French primary school classrooms with learners of English of CEFR A1 level. Quantitative analysis of the interaction data revealed contrasting participation patterns in various task phases and across different areas of the interactional arena. In one class, learners managed the task independently; the teacher intervened only once, and learners exhibited significantly higher on-task time and greater frontstage engagement. In the other class, the teacher participated in backstage task management, providing prompting and echoing, and also in frontstage interaction, and this in all task phases. The study underlines young learners’ capacity for successful L2 interaction in synchronous telecollaboration and traces critical links between learner autonomy and teachers’ interpretation of tasks
Phase-contrast imaging of a dense atomic cloud
No full textInternational audienceWe demonstrate that phase-contrast imaging (PCI) can reliably reconstruct the in situ density profile even for highly spatially and optically dense samples. In our experiment, we achieve high spatial densities of up to 7.9 × 1013 atoms/cm3 and optical depths up to 64 in a dense cold atomic cloud of 88Sr atoms. The use of a spatial light modulator instead of a fixed phase plate in the PCI setup provides enhanced flexibility and control of imaging parameters, making this imaging technique robust against imaging artifacts and adaptable to varying experimental conditions. Moreover, we quantify the conditions under which the standard single-atom polarizability model remains accurate for PCI in a density regime where collective effects should modify the atomic response of the system. We experimentally validate our statements by showing excellent agreement with time-of-flight measurements even at the highest densities. Our results establish PCI as a reliable and versatile method for characterizing spatially dense atomic clouds
Philosophie de l'expérience spectatrice: Puissance de l'art entre science, politique et création de soi
No full textInternational audienceEt si la création ne venait pas de l’artiste, mais du spectateur ? Ce livre avance une thèse radicale : la création étant un acte profondément humain, c’est le spectateur qui en détient les clés. Son désir de connaître, sa nature imitatrice, son intelligence sensible et sa capacité à voir au-delà du visible font de lui un créateur. Ce renversement ouvre une voie nouvelle : celle d’une force créatrice accessible à chacun où toute rencontre avec une œuvre devient l’occasion d’accroître sa puissance d’agir dans le monde. Pourquoi ce livre ?En inscrivant la création dans l’expérience vécue et partagée du spectateur, ce livre propose une autre idée de l’art - mais surtout une autre idée de l’humain, de sa nature et de sa relation au monde.Ce livre est un appel à saisir l’essence créative de nos savoirs, à libérer les forces spontanées d’une véritable pensée du corps, à prendre conscience de la puissance subversive, émancipatrice et transformatrice de l’expérience spectatrice
The Adverse Influence of Maternal Glycaemia During Pregnancy on Offspring's Cardiometabolic Health Profiles
No full textInternational audienceAim: To describe cardiometabolic health profiles at age 5-6 years, their correlation with age at adiposity rebound (AR) and associations with maternal hyperglycaemia at 24-28 weeks' gestation (gestational diabetes, fasting (FPG), 1-hour postload plasma glucose), in mothers without pre-existing diabetes. Methods: BMI, %fat mass, blood pressure (BP), FPG, HOMA-IR and lipids were assessed in children from the EDEN study, a French bicentric birth cohort. Sex-specific cardiometabolic health profiles were derived using principal component analysis and examined against age at AR with Pearson's correlation. Associations with maternal hyperglycaemia were studied using multiple linear regressions adjusted for parental factors. Results: Among 674 children, four profiles were identified per sex: 'higher adiposity, BP, insulin resistance (IR)'; 'higher BP and lower adiposity'; 'higher IR and lower adiposity'; 'higher triglycerides and LDL-c and lower HDL-c'. The profile of 'higher adiposity, BP, IR' was correlated with earlier age at AR in both sexes. Higher maternal FPG was positively associated with the profile of 'higher adiposity, BP, IR' in boys and 'higher triglycerides and LDL-c and lower HDL-c' in girls. Conclusion: 'Higher adiposity, BP, IR' profile at age 5-6 years was associated with earlier age of adiposity rebound. Marginal associations were observed with maternal hyperglycaemia in pregnancy
ClimBurst: A Novel Method to Detect Climatological Anomalies Over Time and Space
No full textInternational audienceDetecting abnormal climate events is crucial for understanding, predicting, and managing climate risks. However, most existing methods require prior knowledge about when and where to search for these events, limiting their effectiveness. In this study, we introduce ClimBurst, a new method to identify climate-related anomalies that does not require any prior information about their duration or spatial extent. We propose computing climate bursts to detect abnormal seasonal activity. The ClimBurst approach can detect anomalies at any time scale. The approach also compares anomalies at neighboring locations enabling the tracking of events across time and space. We apply our method on sea surface temperature data from the Mediterranean Sea between 1960 and 2021, where we detect particularly strong warm anomalies that can last from a few days to a few months over a few kilometers to hundreds, such as the 2015 marine heatwave. Our results reveal a noticeable increase in the frequency, magnitude and the spatial extent of these hot anomalies over time. Researchers and practitioners can use ClimBurst to detect and study climate anomalies, providing a basis for event attribution and long-term trend analysis