HAL Portal Cnam (French National Conservatory for Arts and Crafts)
Not a member yet
31010 research outputs found
Sort by
Système déclaratif et respect de la vie privée, introduction historique
No full textInternational audienc
Filtre adapté normalisé hors-grille : une reparamétrisation robuste en présence de bruit corrélé
No full textInternational audienceUn problème général en radar consiste à détecter si le signal reçu contient ou non un signal d'intérêt spécifique. Nous nous concentrons sur l'un des détecteurs les plus répandus : le filtre adapté normalisé (NMF). Pour être plus réaliste, nous considérons le cas où la cible peut être hors grille. Il existe actuellement différentes méthodes pour effectuer le test NMF hors grille, mais leurs performances ne sont pas robustes, en particulier lorsque le bruit n'est pas blanc. Dans cet article, une nouvelle méthode basée sur une reparamétrisation est proposée. Tout en ayant un coût de calcul similaire aux techniques de l'état de l'art, ses performances de détection sont toujours meilleures
Fault detection and isolation for a class of nonlinear systems based on a bundle of observers and zonotope analysis
No full textInternational audienceThis paper introduces a novel fault detection and isolation (FDI) approach for nonlinear systems subject to unknown but bounded disturbances. The proposed approach combines a bundle of fault detection observers (FDOs), tuned by a peak-to-peak performance technique, with an offline reachability method to generate reliable actuator fault detection and isolation thresholds. Moreover, a sliding-window algorithm, based on zonotopic computation, is designed to be able to provide dynamical fault detection thresholds. This allows one to reduce the conservatism and, by the way, enhance the efficiency of the proposed approach. A quadruple-tank system is considered as a case study, where the theoretical findings of this work are supported by simulation results. In addition, on this example, the performance of the proposed method is compared to that of another method selected from the literature
Contrats de professionnalisation : une coupe budgétaire qui inquiète le secteur de la santé
No full textEnquêter sur les rapports à l’État des gouverné·es: entretien avec Alexis Spire
No full textInternational audienc
Integration of piezoelectric transducers in hydrofoils made of composite materials
No full textInternational audienceBoat appendices, known as hydrofoils, can be subjected to significant dynamic stress during navigation at high Reynolds (Re > 10^5). The turbulent nature of the flow as well as vortex shedding at the trailing edge generate structural vibrations which couples with the resonant modes of the foil. These phenomena, turbulence induced vibration (TIV) and vortex induced vibrations (VIV), leads to undesirable structural fatigue and radiated noise. This study describes the integration of piezoelectric transducers in hydrofoils manufactured in composite material made of carbon fiber and epoxy resin. Macro Fiber Composite (MFC) patches are used to sense and possibly control bending and torsional vibrations induced by the flow. Two hydrofoils with the same external dimensions but obtained with dierent manufacturing processes are considered. The external shape corresponds to a NACA 006 truncated at 80% with a 100 mm chord and 191 mm wingspan. The internal geometries and the arrangement of the composite plies are described for both manufactured hydrofoils. In addition, their respective finite element (FE) models are built according to the manufacturing specifications. Particular attention is paid to the modeling of the piezoelectric transducers and the associated electromechanical coupling. Indeed, the FE models have been created to perform modal analyses of the hydrofoils in air and in water and then optimize coupling factors between the transducer and the structures. To do so, two types of configurations are tested: one with the MFC transducer disabled (short-circuit) and a second with the MFC transducer activated (open-circuit) in order to compute the natural frequencies of the structure in both configurations. The coupling factor, directly related to the dierence between these two natural frequencies, then depends on the location of the MFC on the foil span. The next step in this study is to experimentally validate the coupling factors computed by the FE model. This work is part of the HYDRAVIB project, which aims to develop a hydrofoil vibration mitigation systems based on the integration of piezoelectric transducers
Peut-on jouer à faire des armes? ethno-ergonomie des 24 Heures de l'innovation édition Forces spéciales
No full textInternational audienc
Interpolations spatio-temporelles (4D) basées sur des Graph Neural Network Dynamiques (GNN-D) et applications au stockage géologique de déchets radioactifs
No full textA deep disposal center for radioactive waste, such as Andra's Cigéo project requires continuous, long-term monitoring. This is made possible by a multitude of sensors. However, due to the environmental constraints associated with storage (i.e. radioactivity, the mechanical convergence of galleries, etc.), this network of sensors is prone to degradation over time. Therefore, it is crucial to ensure the consistency of the data collected to guarantee effective monitoring of the center. This means not only identifying sensor failures but also replacing erroneous values with reliable predictions. Graph Neural Networks (GNNs) are suitable tools for these tasks, as they enable accurate representation of system physics and take into account the local topology of the sensor network. In our research, we use data from Andra's underground research laboratory. Specifically, we are using data from an experiment that involved heating a high-activity (HA) cell demonstrator using heating resistors. This setup simulated the heating of radioactive waste on an HA cell. By introducing synthetic errors into this dataset, we trained Graph Neural Networks (GNNs) that leverage both current and historical sensor responses to assess sensor integrity. Additionally, we trained GNNs to generate predictions that could replace the responses of failed sensors, starting from the moment of failure. These GNNs are based on a forward integration mechanism and have been assessed on fundamental thermal simulation problems to evaluate their efficiency and limitations. The architecture of each GNN has been optimized through hyperparameter analysis. Given the large number of variables involved, we proposed a novel method for optimizing GNN architecture based on rating systems. Finally, we compared the performance of the best GNNs with traditional machine learning methods to demonstrate their effectiveness.Un centre de stockage profond de déchêts radioactifs comme le projet Cigéo de l'Andra requiert une surveillance continue sur le long terme. Cette surveillance est possible grâce à une multitudes de capteurs. Cependant, en raison des contraintes environementales associées au stockage (radioactivité, convergence mécanique des galeries, etc.) ce réseau de capteurs est sujet à une dégradation au fil du temps. Il est alors crucial d'assurer la cohérence des données recueillies pour garantir la surveillance du centre. Pour ce faire, il faut non seulement identifier les défaillances de capteurs mais aussi remplacer les valeurs erronées par des prédictions cohérentes. Les Graph Neural Networks (GNN) sont des outils adaptés pour ces tâches car ils permettent de représenter précisément la physique du système et prennent en compte la topologie locale du réseau de capteurs. Dans nos travaux, nous utilisons des données issues du laboratoire de recherche souterrain de l'Andra. En particulier, celles d'une expérience de chauffe d'un démonstrateur de cellule haute-activité (HA) par des résistances chauffantes. Ce qui permet d'imiter la chauffe d'une alvéole HA par des déchêts radioactifs. En ajoutant des erreurs synthétiques à ces données, nous avons entrainé des GNN qui utilisent les réponses capteurs (présentes et passées) pour détecter les défaillances capteurs. A l'aide des mêmes données, nous avons entrainés des GNN effectuant une prédiction au niveau des capteurs défaillants, à partir de l'instant de la panne. Ces GNN se basent sur un mécanisme d'intégration temporelle et ont été étudiés sur des problèmes de simulation thermique élémentaires afin d'évaluer leur efficacité ainsi que leurs limites. L'architecture de chacun des GNN a été optimisée par le biais d'un analyse hyper-paramétrique. En raison du grand nombre de variables, nous avons proposé une nouvelle méthode d'optimisation de l'architecture des GNN basée sur les systèmes de classement. Enfin, nous avons comparé les meilleurs GNN à des méthodes d'apprentissage machine traditionnelles afin de prouver leur efficacité
Could the rise in poultry production be driving an increase in Human Salmonella enterica subsp. arizonae infections?
No full textInternational audienceBackground: Salmonellosis represents the second leading cause of bacterial foodborne illness in Europe. Among all species, Salmonella enterica is the most pathogenic, with more than 2600 serotypes, and has the capacity to infect humans, animals, plants, and the environment. The subspecies enterica (I) is typically found in livestock such as poultry, pigs and cattle, while the subspecies arizonae (IIIa) is mainly associated with cold-blooded animals. However, since 2018, there has been a noticeable increase in human cases of Salmonella enterica subsp. arizonae (serotype: 48:z4,z23:-) with no clear source identified. This rise, which poses a significant risk to humans health, is also being observed in poultry production. The aim of this study is to investigate, through whole genome sequencing (WGS), whether there is a phylogenetic link between poultry and human strains. Methods: For this purpose, available human genomes from the National Reference Center for Salmonella at the Pasteur Institute (CNR-ESS) and poultry genomes from the National Reference Laboratory (NRL) for Salmonella database were analyzed using whole genome sequencing (WGS). The strains were compared through core genome multi-locus sequence typing (cgMLST) and Hierarchical Clustering (HC5) as provided by EnteroBase database (www.enterobase.warwick.ac.uk). Results and discussion: The results confirmed the emergence of Salmonella enterica subsp. arizonae serotype (48:z4,z23:-) in France, as evidence by the presence of distinct strains in both human and poultry populations. This study revealed significant genetic similarities between the strains isolated from human cases and those found in poultry farms, suggesting potential epidemiological links between the two. These findings raise significant concerns about the potential transmission of the pathogen from animals to humans, emphasizing the need for continuous surveillance of Salmonella spp. in the poultry industry, given its critical role in the zoonotic transmission of this pathogen. This study also highlights the importance of epidemiological investigations using WGS to better understand how this pathogen is transmitted to humans