Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
SENSE-LM : une synergie entre modèles de langage et représentations sensorimotrices pour la recherche de références olfactives et auditives dans des documents écrits
No full textInternational audienceLes cinq sens – la vue, le goût, l'odorat, l'ouïe et le toucher – influencent la perception humaine à travers diverses modalités. L'extraction de références sensorielles (expressions évoquant une expérience sensorielle) dans un corpus textuel est un sujet d'intérêt avec de nombreuses applications, notamment en neurosciences ou dans l'étude du patrimoine culturel. Cet article présente SENSE-LM, un système d'extraction d'informations sensorielles dans de grandes collections de documents textuels. En associant des modèles de langage et des représentations sensorimotrices, il automatise l'extraction d'informations sensorielles à gros grain (classification binaire de phrases) et à grain fin (extraction de termes sensoriels). L'évaluation de SENSE-LM sur deux sensorialités, l'odorat et l'ouïe, révèle des avancées significatives dans l'automatisation de ces tâches complexes
An analysis of the effect of the jet initial conditions on the wavelet separated near-field acoustic pressure
No full textInternational audienceThis paper reports a parametric investigation of the effect of the nozzle exhaust initial conditions on the wavelet separated acoustic pressure generated by a single stream compressible jet in its near field from a database obtained by Large-Eddy Simulations of jet flows at M = 0.9 and Re = 10 5 . The nozzle-exit boundary-layer conditions consist of different turbulence intensities for fixed thickness and several thicknesses in laminar conditions. Pressure time series are extracted from virtual probes distributed in the near field of the jets and the acoustic components of the near field pressure are extracted using a wavelet-based procedure able to decontaminate the signals from the hydrodynamic contribution. The reconstructed acoustic time series are analyzed in the frequency domain and in terms of Overall Sound Pressure Level (OASPL). The results show that both the boundary-layer thickness and the turbulence level significantly affect the acoustic pressure in terms of both intensity and directivity. In the laminar case, strong sideline components are observed and strongly depend on the boundary layer thickness. These components clearly appearing in the energy spectra are associated with the Kelvin-Helmholtz instability waves. For large nozzle-exit turbulence intensities, the acoustic field is more uniform and less intense in the sideline direction. On the other hand, the streamwise directivity of the acoustic pressure appears to be strictly correlated to the length of the jet potential core which strongly varies with the initial conditions
Investigation of waste materials and application for bio-batteries system proposals for low-power devices
No full textGenerating energy from biowaste has been at the helm of research given both its potential to aid in the decarbonization process and its renewable nature. However, the techniques used remain complex and non-portable, slowing down the adoption of bioenergy generation. Microbial fuel cell (MFC) is a technology that generates electrical energy directly from biodegradable waste. This technology utilizes the principle of conventional electrical cells. The microorganisms present in the anodic chamber break down the waste and generate electrons and protons. The electrons attach to the anode electrode to form a biofilm, while the protons pass through the membrane to join the cathodic compartment where an oxidant and the cathode electrode are present. Once a conductive wire connects the two electrodes together, the electrons begin to flow to join the protons present in the cathodic chamber, and this movement of charge carriers generates electricity. This method of electricity production is increasingly attracting researchers, and efforts made with the technology are still primarily limited to the laboratory scale, restricting its interest and utilization for electrical power generation. It is important to find ways of increasing the electrical power production of MFC in order to cover applications such as supplying low-power devices like those usually used in Wireless Sensor Networks (WSN), and lighting in remote areas of developing countries where local communities don’t have access to conventional electricity. The main objective of this research project is to propose a high-performance, cost-effective, and stable MFC energy production system with a simple manufacturing process. This work first focuses on the investigation into the potential of excreta (animal and human) to produce electricity through microbial fuel cell technology, while considering the factors influencing the production yield of MFC in order to improve the production and achieve optimal results. Two types of waste are specifically chosen for this purpose: cow dung and human faeces. The reactors (vessels used to carry out chemical reactions) are 3D printed using polylactic acid (PLA) material to ensure similarity, sealing, and reproducibility of results. After various characterizations, cow dung showed a better potential for electricity production through microbial fuel cell technology compared to human faeces. To afford the required energy, MFCs are put in a stack; cow dung is chosen and used for experimentation to investigate different configurations such as series, parallel, series-parallel combination, and parallel-series. The conduct of these investigations will highlight the configuration that can minimize voltage reversal, which is responsible for losses that drop down the production efficiency of MFCs when stacked together. The results show that the parallel-series combination limits losses in the stack assembly while increasing the total power of the system. This configuration can be used for small-scale MFCs to minimize losses and circumvent the use of protection circuits designed to prevent voltage return in MFC systems. To limit the costs associated with the implementation of MFCs and encourage local large-scale production of stacks, the use of concrete is investigated for the fabrication of MFC reactors. With those concrete-based reactors, in addition to demonstrating their feasibility, we achieved a maximum power density of 20.21 mW/m², which is higher than the 14.1 mW/m² obtained with the 3D-printed reactors from the first experiment. Furthermore, the production of a dual-chamber concrete-based reactor here is 5.17 times less expensive than that of the 3D-printed single chamber made with PLA plastic filament.La production d'énergie à partir de biodéchets a été à la tête de la recherche, compte tenu à la fois de son potentiel de contribution au processus de décarbonisation et de sa nature renouvelable. Cependant, les techniques utilisées restent complexes et non portables, ce qui ralentit l'adoption de la production de bioénergie. La pile à combustible microbienne (PACM) ou biopile est une technologie qui génère de l'énergie électrique directement à partir de déchets biodégradables. Cette technologie utilise le principe des cellules électriques conventionnelles. Les micro-organismes présents dans la chambre anodique décomposent les déchets et génèrent des électrons et des protons. Les électrons se fixent à l'électrode d'anode pour former un biofilm, tandis que les protons traversent la membrane pour rejoindre le compartiment cathodique où un oxydant et l'électrode cathodique sont présents. Une fois qu'un fil conducteur relie les deux électrodes ensemble, les électrons commencent à circuler pour rejoindre les protons présents dans la chambre cathodique, et ce mouvement des porteurs de charge génère de l'électricité. Cette méthode de production d'électricité attire de plus en plus les chercheurs, et les efforts déployés avec cette technologie restent principalement limités à l'échelle du laboratoire, restreignant ainsi son intérêt et son utilisation pour la génération d'énergie électrique. Il est important de trouver des moyens d'augmenter la production d'énergie de la PACM afin de couvrir des applications telles que l'alimentation d'appareils à faible puissance, comme ceux généralement utilisés dans les réseaux de capteurs sans fil (WSN), et l'éclairage dans les zones reculées des pays en développement où les communautés locales n'ont pas accès à l'électricité conventionnelle. L'objectif principal de ce projet de recherche est de proposer un système de production d'énergie de PACM performant, économique et stable, avec un processus de fabrication simple. Ce travail se concentre d'abord sur l'étude du potentiel des excréments (animaux et humains) pour produire de l'électricité grâce à la technologie des piles à combustible microbiennes, tout en tenant compte des facteurs influençant le rendement de production des PACMs afin d'améliorer la production et d'obtenir des résultats optimaux. Deux types de déchets sont spécifiquement choisis à cet effet : la bouse de vache et les excréments humains. Les réacteurs (cuves utilisées pour réaliser des réactions chimiques) sont imprimés en 3D à l'aide de matériau en acide polylactique (PLA) pour garantir la similitude, l'étanchéité et la reproductibilité des résultats. Après diverses caractérisations, la bouse de vache a montré un meilleur potentiel de production d'électricité à travers la technologie des PACMs par rapport aux excréments humains. Pour augmenter l'énergie produite par les PACMs, des configurations en « stack » sont expérimentées avec de la bouse de vache. L'étude examine différentes configurations (série, parallèle, série-parallèle, parallèle-série) pour minimiser l'inversion de tension, source de pertes d'efficacité. Les résultats indiquent que la configuration parallèle-série réduit les pertes et augmente la puissance totale, ce qui la rend adaptée aux PACMs à petite échelle, tout en évitant l'utilisation de circuits de protection contre le retour de tension. Pour réduire les coûts d'implémentation et favoriser la production locale des PACMs, des réacteurs en béton sont utilisés. Ces réacteurs ont montré leur faisabilité avec une densité de puissance maximale de 20,21 mW/m², surpassant les 14,1 mW/m² des réacteurs imprimés en 3D de la première expérience. De plus, la fabrication d'un réacteur à béton à double chambre est 5,17 fois moins coûteuse que celle d'un réacteur à chambre unique en plastique PLA
Recent advances in film dosimetry for quality assurance in microbeam radiation therapy
No full textInternational audienceMicrobeam Radiation Therapy (MRT) is currently on the verge of entering the clinical phase, necessitating robust quality assurance (QA) with appropriate dosimeters to ensure patient safety. However, MRT dosimetry presents significant challenges due to its intrinsic characteristics, including steep dose gradients with alternating high and low dose regions—referred to as peaks and valleys (valley doses comprising only 5–7 of peak doses)—micrometric resolution requirements, the use of low-to-medium energy X-rays and extremely high dose rates of up to 16 kGy/s. Currently, Gafchromic films are considered the gold standard for patient QA. Previous studies have reported discrepancies of 10–30 between measured and calculated peak and valley doses. This study aimed to evaluate four types of films (EBT-3, EBT-XD, MD-V3, and HD-V2) to determine the most suitable option for MRT QA. To support dosimetric benchmarking, the microDiamond detector was used as the reference for MRT measurements, and the ionization chamber for broad beam reference dosimetry. Discrepancies were observed between calculated and film-measured peak and valley doses, and several hypotheses were proposed to explain these differences. Possible causes include simplifications in dose calculation engines, intrinsic limitations of the films, uncertainties in experimental setups, and variations in readout methods. The study suggests that MD-V3 may be the most suitable film for MRT dosimetry, offering a dynamic range that effectively covers both peak and valley doses
A Multimodal Evaluation Pipeline for Mathematical Expression Recognition: Comparisons of Datasets, Metrics, and Models
No full textInternational audienceSince OpenAI's 2016 Request for Research on Image to LaTeX, printed mathematical expression recognition has advanced significantly, with newer techniques like Vision Transformers further enhancing accuracy. However, evaluations often remain limited to a small set of datasets and primarily focus on LaTeX string similarity, neglecting semantic and visual aspects. To perform a broader evaluation, we benchmarked five models representative of the evolution of Vision Transformers across eight datasets over multiple modalities. Our proposed multimodal evaluation pipeline, PiE-MER, converts predicted LaTeX into MathML, Label Graphs, and normalized images. This range of representations allows us to compute additional metrics that better capture models' syntactic, semantic, and visual accuracy.Correlation analysis revealed that BLEU and Levenshtein alone are insufficient for complete evaluations. We propose complementary metrics, including image-based Levenshtein, MathML-based scores, and graph evaluation. The MathNet model achieved the highest average performance across the datasets, with a 87% LaTeX edit score, thanks to its Convolutional Vision Transformers and multi-font training dataset. Our pipeline and benchmark can offer a clearer and more robust evaluation of future recognition systems. A GitHub repository containing the proposed evaluation pipeline is available at https://github.com/fwieckowiak/PiE-MER, enabling future authors to evaluate their own models and datasets
A Fuzzy Logic-Based System for Detecting Trustable Physical Unclonable Functions
No full textInternational audiencePhysical Unclonable Functions (PUFs) provide a promising security mechanism by leveraging inherent process variations to generate unique, hardware-bound secrets without requiring secure storage. However, ensuring PUF reliability and detecting potential tampering remain critical challenges. This paper presents a fuzzy logic-based classification system that determines the authenticity of PUF responses using three key metrics: Reliability, Stability, and Reliability Invariance. The system classifies PUF responses into three categories: Trustable, Tampered, and Undecided. This approach enhances the automatic detection of unreliable responses that may indicate tampering while ensuring the fidelity of PUF responses over time. By applying fuzzy inference rules, our method achieves high accuracy in distinguishing between trustworthy and compromised PUFs. Experimental results demonstrate the effectiveness of our approach, making it a valuable method and tool for hardware security applications
Rx-PAD : Recognition and eXtraction - a dataset for Prescription Analysis and clinical Data structuring
No full textInternational audiencehis paper introduces a novel dataset designed to facilitatethe deep structuring of medical prescriptions, addressing the critical needfor accurate and efficient information extraction from semi-structured,sensitive patient health scanned documents. By structuring these pre-scriptions, medical professionals can more effectively verify dosage in-structions and ensure the correct delivery of prescribed drugs to pa-tients. Unlike existing document analysis datasets, Rx-PAD dataset isspecifically tailored to the complex syntax and semantic relationships ofmedical prescriptions. Rx-PAD consists of 200 fully annotated imagescollected from French pharmacies, supporting two primary tasks: EntityExtraction (EE, 61 labels) and Entity Linking (EL, 11 relation types).Each token is tagged to infer entities and group IDs, enabling a detailedlevel of structured clinical instruction. To evaluate the efficiency of au-tomatic prescription structuring, we propose a baseline implementationthat is both efficient and precise in detecting medicines and dosages. Toevaluate model performance in a real-world medical context, we intro-duce the Drug Accuracy and Completeness (DAC) metric, assessing thecorrect linking of drug names, dosings, and forms. Rx-PAD provides arobust foundation for developing solutions that can handle sensitive med-ical data with the accuracy and speed required for real-world applicationssuch as pharmacy automation, medical AI, and healthcare complianceapplications, contributing to meaningful healthcare advancements. Thedataset will be publicly available to foster further research in medicaldocument understandin
Neural Field Equations and Hawkes processes: long-term stability of traveling wave profiles in the neutral case
No full textWe consider the long-time behavior of a population of interacting Hawkes processes on the real line, with spatial extension. The large population behavior of the system is governed by the standard Voltage-based Neural Field Equation (NFE). We prove long-time stability of the system w.r.t. traveling wave solutions for the NFE on a diffusive time scale, in the neutral case, that is when the speed of the traveling wave solution is zero: the position of the traveling wave profile becomes essentially Brownian on a diffusive time scale. We borrow here from the seminal work of Chevallier, Duarte, L\"ocherbach and Ost concerning the approximation of NFE by Hawkes processes, the existence result of traveling waves of Ermentrout and McLeod and from the stability result of the traveling wave profile from Lang and Stannat. We provide also similar results concerning the activity based NFE
A new reduction strategy for direct time integration of cyclic symmetric systems featuring unilateral contacts
No full textInternational audienceDirect time integration of full cyclic systems is frequently prohibitive due to the presence numerous unknowns. In order to deal with this issue, cyclic symmetry is generally used in a linear framework to reduce the problem but it generally does not apply to nonlinear systems. Nevertheless, analyzing the nonlinear forces of a specific problem can still yield a reduction. This contribution proposes a reduction technique to the nonlinear equations of motion allowing for direct integration of systems in cyclic symmetry. The strategy relies on the equivalence between physical and spectral spaces, linking sectors and nodal diameters. This methodology is applied to a simplified model featuring unilateral contacts. Several contact scenarios are studied to demonstrate that the reduction strategy is able to compute the response of the full system by only integrating the system over a subset of its total sectors, allowing to achieve high computational time saves
Reduction of cyclic symmetric mechanical systems subject to unilateral contact: Application to periodic blade-tip/casing contact interactions
No full textInternational audienceThis paper proposes a novel reduction technique for industrial bladed disks featuring blade-tip/casing contacts. By analyzing a regularized contact law, a criterion allowing to drastically reduce the size of the problem is derived. It consists in restricting solutions to only a subset of nodal diameters identified in advance. An additional parity rule allowing to reduce the size of problem by half in certain configurations when solving through the harmonic balance method (HBM). These rules are tested on an academic cyclic model featuring unilateral contacts. The reduction is exact both for a regularized law but also for a nonsmooth contact law. The results match perfectly with the reference calculations led on the full system. All the previously mentioned observations remain true for the simulations led on the industrial fan stage ECL5/Catana under rubbing interactions. After including friction forces, the reduction technique is still exact for the industrial model. The previously published traveling wave hypothesis (deducing the full response from a single sector) is shown to perform well for purely traveling excitations. Several loading scenarios demonstrate the applicability of the strategy in numerous configurations. Localized responses are accurately captured by the reduction method in the case of excitation through perturbed traveling waves. This contribution allows to take a major step forward in the numerical modeling of bladed disks featuring blade-tip/casing interactions by taking advantage of cyclic symmetry. The proposed strategy yields an average reduction of 37.6% for the HBM and 25.2% for other methods. For traveling wave excitations, the reduction is rapidly over 90% and goes up as the number of sectors increases