Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Can we predict the magnetic loss of a nanocrystalline core from LCR-meter characterization?
No full textInternational audienceNanocrystalline cores are key for high-frequency applications due to their low hysteresis losses, but predicting dynamic losses remains challenging. This study explores using LCR-meter-based permeability spectroscopy (μS) to characterize these materials and predict losses via fractional derivative models. The method offers a rapid, industry-friendly alternative to complex hysteresis setups.</div
Hybrid Electrokinetic-Plasmonic Biosensor for Fast and Sensisitve Pathogen Detection in Irrigation Water
No full textInternational audienc
Contribution of curvature to cell division and renal tissue organization
No full textInternational audienc
Integration in finite terms and exponentially algebraic functions
No full textWe develop techniques at the interface between differential algebra and model theory to study the following problems of exponential algebraicity: (1) Does a given algebraic differential equation admits an exponentially algebraic solution, that is, a holomorphic solution which is definable in the structure of restricted elementary functions? (2) Do solutions of a given list of algebraic differential equations share a nontrivial exponentially algebraic relation, that is, a nontrivial relation definable in the structure of restricted elementary functions? These problems can be traced back to the work of Abel and Liouville on the problem of integration in finite terms. This article concerns generalizations of their techniques adapted to the study of exponential transcendence and independence problems for more general systems of differential equations.As concrete applications, we obtain exponential transcendence and independence statements for several classical functions: the error function, the Bessel functions, indefinite integrals of algebraic expressions involving Lambert's W -function, the equation of the pendulum, as well as corresponding decidability results
State of the art on the sustainability of lithium-ion batteries for electric mobility
No full textInternational audienceThe massive electrification of vehicles is a key point in the energy transition. This shift leads to various challenges, especially regarding lithium-ion batteries, as they can cause several environmental and socioeconomic impacts throughout their lifecycle. Therefore, it is essential to assess these effects in order to reduce them. This paper provides a comprehensive study of the literature regarding lithiumion batteries sustainability (ecological, social and economic aspects) in the context of electric mobility. This paper highlights the growing number of published Life Cycle Analyses (LCA) addressing environmental impacts, while emphasizing the need for methodological harmonization at multiple levels, particularly in LCA tools and methods, as well as in the battery life cycle models considered. The lack of transparent and accessible data has been also highlighted. In the case of Social Life Cycle Analysis (SLCA), evaluating social impacts, a significant lack of studies and data have been noted. Moreover, this methodology requires further clarification and development. Life Cycle Costing (LCC), assessing economic impacts, rarely covers the entire life cycle of batteries, and its framework needs to be defined more clearly. Finally, sustainability studies, combining the three previous concepts, are rare, and the methodological frameworks and links between the three sustainability dimensions need to be clarified. This work explicitly identifies the obstacles and levers for accurately assessing the sustainability of lithium-ion batteries. In particular, recommendations are made on three major points. Firstly, the harmonisation of assessment methodologies. Secondly, the need for interdisciplinary contributions to develop robust LCA models, illustrated by the example of the electrical engineering community, which can contribute to the integration of behavioural and ageing models, usage scenarios, as well as account for the diversity of battery technologies. Thirdly, recommendations are made to support the development of more robust data and models, through open-science or the development of a secure data-sharing framework. Finally, all the data extracted from our study are open access.</div
Combining X-ray tomography and digital volume correlation to measure 3D bulk mechanical fields in Hertzian contact mechanics
No full textIn this study, we revisit classical Hertzian contact theory through a novel experimental approach that enables in situ 3D imaging of contact mechanics using laboratory-based X-ray tomography. By employing a model contact system—a PDMS sphere indented into a PMMA plane—we measure the 3D displacement field within the deforming elastomeric PDMS sample during indentation using digital volume correlation (DVC).We introduce multiple methods to experimentally evaluate the stress field beneath the contact surface and directly compare it to the predicted Hertzian stress distribution. Our approach leverages DVC-based finite element simulations, where boundary conditions are fully derived from the experimentally measured displacement field. This eliminates reliance on contact assumptions or numerical contact algorithms, ensuring that the stress field is determined solely by experimental data.To our knowledge, this work presents the first experimentally driven characterization of a full-field stress distribution beneath the contact surface, demonstrating qualitative agreement with classical contact models. However, our findings also indicate that achieving quantitative agreement will require the development of more complex models, which do not currently exist, as well as advancements in DVC regularization schemes to account for surface-specific behaviors.While this paper focuses on normal loading conditions, the experimental setup and methodology are fully adaptable to any type of contact loading scenario
Approximation efficace pour l'explication locale des modèles d'apprentissage, appliquée à l'inférence d'activité locale des réseaux de régulation génique
No full textThe work presented in this thesis is divided between algorithm design for machine learning explainability and single-cell RNA-Seq data analysis. There has been a dialogue between the two aspects of the work that guided the research. The project was initiated with the goal of helping to understand the difference in behavior between populations of cells, by focusing on regulation of gene expression. In particular, we are interested in changes in the regulation of genes, modeled as rewiring events in Gene Regulatory Networks (GRNs). This initial question emerged from a collaboration with the Centre de Recherche en Cancérologie de Lyon (CRCL). For this study, we trained machine learning models to predict gene expression as a function of the expression of transcription factors, which are genes that regulate the expression of other genes. We then applied a machine learning explainability method that allowed us to calculate an importance score associated with each of the model parameters. This importance is computed locally, for one prediction associated to one instance (cell), which makes it possible to leverage single-cell resolution and capture cellular heterogeneity. We focused on the SHapley Additive exPlanation (SHAP) framework to provide importance scores. This method only applies the model on inputs and therefore is compatible with any type of predictive model: it is model-agnostic. Computing SHAP values presented a significant challenge because its cost is exponential with respect to the number of features. However, in the context of feature importance computation for the inference of regulatory relationships, we argued that only a subset of the features of the model are truly impactful. Therefore, we strategically directed the computational resources towards computing the SHAP values for the top-k most important features. To reduce the cost of computing the top-k SHAP values, we designed and implemented TopShap, an iterative algorithm that interleaves refinements of the SHAP value approximation with pruning steps to discard elements that can no longer be in the top-k. We showed that TopShap is faster than post-processing the output of the fastest model-agnostic approximation method: Kernel SHAP. Next, we used TopShap to study GRN rewiring events. This led to the design of a workflow named re_actShap, which was then extended and applied to cancer data provided by collaborators at CRCL.Le travail présenté dans cette thèse se divise entre la conception d'algorithmes pour l'explicabilité de l'apprentissage automatique (Machine Learning, ML) et l'analyse de données de séquençage d'ARN unicellulaire (single-cell RNA-sequencing). Un dialogue s'est instauré entre les deux aspects du travail qui a guidé la recherche. Le projet a été initié avec l'objectif d'aider à comprendre la différence de comportement entre les populations de cellules, en se focalisant sur la régulation de l'expression des gènes. Nous nous sommes particulièrement intéressés aux changements dans la régulation des gènes, modélisés comme des rebranchements dans les réseaux de régulation de l'expression génique. Notre question initial est issue de la collaboration avec le Centre de Recherche en Cancérologie de Lyon (CRCL). Pour cette étude, nous entraînons de modèles d'apprentissage automatiques à prédire l'expression des gènes en fonction de l'expression des facteurs de transcription, qui sont les gènes qui permettent la régulation de l'expression des autres gènes. Nous appliquons ensuite une méthode d'explicabilité de l'apprentissage automatique qui permet de calculer un score d'importance associé à chacun des paramètres du modèle. Cette importance est calculée localement, pour une prédiction associée à une instance (cellule), ce qui permet de tirer parti de la résolution unicellulaire offerte par les données de single-cell RNA-sequencing et capturer l'hétérogénéité cellulaire. Nous avons choisi d'utiliser pour cela le framework "SHapley Additive exPlanation" (SHAP), qui retourne des scores d'importances. Cette méthode nécessite seulement d'appliquer le modèle à de nombreuses instances, et est donc compatible avec tout type de modèle prédictif : elle est indépendante du modèle. Le calcul des valeurs SHAP présente un défi important, car sa complexité est exponentielle par rapport au nombre de paramètres. Cependant, dans le contexte de la détection de la régulation des gènes, nous pensons que seul un sous-ensemble des paramètres du modèle est important. Par conséquent, nous avons stratégiquement orienté les ressources de calcul vers le calcul des k plus importantes valeurs SHAP. Pour réduire le coût de calcul des k plus importantes valeurs SHAP, nous avons conçu et implémenté TopShap, un algorithme itératif qui fait intervenir de concert l'approximation des valeurs SHAP et le retrait des valeurs qui ne sont pas candidates à faire partie des k plus importantes valeurs SHAP. Nous avons démontré que TopShap est plus rapide que le post-traitement de la sortie de la méthode d'approximation agnostique du modèle la plus rapide : Kernel SHAP. Nous l'avons ensuite utilisé pour étudier les évènements de rebranchements dans les GRN. Cela a conduit à la conception d'un workflow nommé re_actShap, ensuite appliqué aux données fournies par des collaborateurs du CRCL. En plus de la détection des rebranchements dans les GRNs à l'aide de TopShap, d'autres analyses ont été menées sur cet ensemble de données, ce qui a permis de formuler deux hypothèses qui ont été testées expérimentalement
Graph Consistency Rule Mining with LLMs: an Exploratory Study
No full textInternational audienceGraph data structures are essential for representing complex relationships in various domains, including life sciences, social media,healthcare, finance, security, and planning. With the increasing reliance on graph databases—particularly property graphs—for capturing semantic relationships, ensuring data integrity and quality has become crucial. Traditional methods for maintaining consistency, such as expert-defined rules and data-mined constraints like functional and entity dependencies, face challenges in scalability, adaptability, and comprehensibility. In this paper, we explore how Large Language Models (LLMs) can be utilized to automatically generate and refine consistency rules for property graphs through guided prompts. Leveraging the reasoning capabilities of LLMs over expressive graph models, we conduct an exploratory empirical study to assess the extent to which LLMs can extract rules that enforce data consistency. Our evaluation spans different real-world datasets and various graph encoding methods. Our results demonstrate that LLMs show promising abilities in extracting consistency rules, primarily identifying schema-based constraints such as primary keys, attribute uniqueness, and label enforcement. Additionally, LLMs occasionally capture more complex patterns, including temporal constraints where certain events cannot occur simultaneously
Permanent degradation of p-GaN HEMTs due to repetitive overvoltage stress during hard turn-off switching
No full text#58International audienceGallium Nitride High Electron Mobility Transistors (GaN HEMTs) are promising candidates for power electronicsbut suffer from reliability concerns due to, for instance, their lack of avalanche capability. This study investigatesthe long-term impact of dynamic overvoltage stress on GaN HEMTs using a newly designed test circuit, UIS3,which isolates key stress factors. Devices were subjected to short-duration repetitive overvoltage stress near theirdynamic breakdown voltage. Characterization before and after stress reveals permanent degradation in CDS andIDSS , suggesting structural damage within the device. A distinct alteration in the CDS curve is observed, possiblylinked to local breakdown in the insulating layer. Higher stress levels accelerate failure. Our results provide newinsights into GaN HEMT degradation mechanisms under high-voltage stres
Fabrication and Optimization of Type II Silicon Clathrate Films
No full textInternational audienceSilicon clathrates (SiCL) are a class of cage-structured materials with tuneable optoelectronic properties, offering significant potential for photovoltaic applications. Among them, type II SiCL (NaxSi136) are particularly attractive due to their unique ability to reversibly accommodate guest atoms without compromising the crystal structure. Traditional synthesis methods, such as synthesis under extreme pressure or fabrication under glove boxes, although essential for probing the intrinsic properties of clathrates under well-controlled conditions, are not well-suited for scaling up and are often energy-intensive and require specialized equipment. To overcome these limitations, a scalable, glovebox-free synthesis method based on two thermal decomposition steps was developed to produce silicon clathrate films. Post-synthesis treatment, including thermal pressing and reactive ion etching, was employed to enhance the electrical properties of the material. Phase formation and crystallinity are confirmed by X-ray diffraction and Raman spectroscopy; morphology is assessed by scanning electron microscopy (SEM); and optoelectronic properties are evaluated by photoluminescence. This approach provides a reproducible and scalable route to fabricate type II silicon clathrate films, with promising potential for integration into optoelectronic devices