Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
L’ethnologie debout ! Les ficelles de l’enquête en milieu militaire
No full textInternational audienceLe travail de recherche repose largement sur l’autonomie des chercheurs. En matière de choix de terrain, de productionet d’analyses de données, répondre aux demandes d’un prescripteur ou se soumettre à sa censure réduit considérablementles chances de produire de la « bonne » science sociale. Pourtant, il est des espaces et des activités qui ne peuvent pas,d’emblée, être transformés en terrain de recherche, sans ne risquer que les savoirs alors produits exposent les acteursétudiés à des périls parfois vitaux. Les temps et les lieux du travail combattant sont de ceux-là. Le présent article proposeainsi une méthode pour en « conventionnaliser » l’étude, de telle sorte qu’il soit possible aux ethnographes de produireune « bonne » science sociale du « faire » guerrier sans pour autant que les connaissances publiées n’attentent à la vie dessoldats. De manière plus générale, cet article critique toute propension à l’exclusivisme méthodologique et plaide pourune combinatoire des formes d’enquête
Neural Data Assimilation for Regime Shift Monitoring of an Idealized AMOC Chaotic Model
No full textInternational audienceData assimilation (DA) reconstructs and forecasts the dynamics of geophysical processes using available observations and physical a priori. Recently, the hybridization of DA and deep learning has opened new perspectives to address model-data interactions. This paper explores its potential contribution to the analysis of a chaotic oceanic phenomenon: the centennial to millennial variability of the North Atlantic ocean circulation during the last glacial period. The implemented neural approach-4DVarNet-yields meaningful improvements over a classical variational DA method in reconstructing regime shifts of the Atlantic Meridional Overturning Circulation (AMOC), especially when fewer observations are available. Interestingly, results exhibit that exploiting explicitly the a priori dynamical model does not lead to better performances compared to a data-driven model. Additionally, we compare four sampling strategies to assess how observation patterns influence the capture of unstable AMOC phases. We highlight the gain of regular over random sampling strategies, with reconstruction errors dropping below 2% for a 100-year sampling period. We find that monitoring the AMOC with regular clusters of three consecutive observation points can reduce errors by a factor of five. Eventually, we assess 4DVarNet's robustness in reconstructing a partially-observed system and in generalizing to different dynamical regimes. We also investigate on the maximum sampling periods that 4DVarNet can assimilate without compromising reconstruction quality. This study, based on an idealized yet complex physical model, suggests that neural approaches trained on observations wisely acquired could improve the monitoring of regime shifts in the context of climate change
Multicriteria File-Level Placement Policy for HPC Storage
No full textInternational audienceThe rapid expansion of data volumes across various scientific and technical fields, along with the development of exascale computing in the high performance computing (HPC) domain, continually challenge existing storage systems. These systems typically consist of heterogeneous multi-tier storage architectures, ranging from high-speed solid-state drives (SSDs) tier with limited storage capacity to slower magnetic tapes tier with larger storage capacity. A significant challenge in HPC storage systems is the effective placement and migration of data across different storage levels. Current strategies, such as those implemented in parallel file systems like Lustre, utilize hierarchical storage management (HSM) solutions such as the Robinhood Policy Engine, which operate at the file granularity level for data eviction policies. In contrast, traditional caching policies work at the block level. This mismatch of granularity makes it difficult to adopt traditional eviction policies to those HSM. This study introduces a new multi-criteria file-level eviction policy incorporating frequency and recency of access, file lifetime, and a fairness criterion. Our policy reduces I/O processing times by average of 30% for tested workloads and improves the hit ratio by 56.43% on average, outperforming block-based cache replacement policies such as LRU, LFU, and ARC
Laguerre–Gaussian laser filamentation in ambient air
No full textInternational audienceThe filamentation of ultrashort laser pulses in air with Laguerre–Gaussian beams opens up numerous possibilities for atmospheric applications. In this work, we demonstrate a novel method to measure the critical power for self-focusing of structured laser beam based on the acoustic detection of laser ionization. Using this method, we determine the critical power of different Laguerre–Gaussian beams at 800 nm with ultrashort (50 fs) and sub-picosecond (500 fs) pulse durations. We also discuss the effect of the numerical aperture of the beam on the critical power and present the first measurements of the energy deposited by Laguerre–Gaussian filaments in the air. Our results reveal an unexpected influence of the laser pulse duration on the filamentation of vortex beams
Efficient Hardware Primitives for Interval Contractors in Robotics and Integration to a Custom RISC-V ISA Extension
No full textInternational audienceInterval analysis is commonly used in mobile robotics to solve nonlinear problems. Typical implementations mostly rely on software libraries which perform poorly on low-power embedded devices. This article shows how interval primitives can be ported on FPGA and integrated to the RISC-V architecture using a custom ISA extension called xinterval. The novelty of this approach is the emphasis on elementary interval contractors which are higher-level abstractions than plain interval arithmetic especially well suited for robotics applications
Challenges and Performance of SLAM Algorithms on Resource-constrained Devices
No full textInternational audienceEvaluating the performance of Simultaneous Localization and Mapping(SLAM) algorithms is essential for the progress of robotic systems. However,conducting a comprehensive assessment of SLAM systems in the context ofrecent advancements is challenging due to the wide variety of hardware plat-forms, algorithm configurations, and datasets available. This study aims totest SLAM algorithms on resource-constrained devices such as the NVIDIAJetson AGX Orin 64GB. Experiments are conducted with various visual-based localization algorithms that either leverage deep learning models forspecific tasks within the SLAM process or are learned end-to-end to estimatecamera pose. The evaluation focuses on the following systems: RDS-SLAMand VDO-SLAM, which utilize semantic information to achieve precisemotion estimation; TSformer-VO, an end-to-end Transformer-based modeldesigned for monocular visual odometry; and DeepVO, which based onrecurrent neural networks. The systems are evaluated using several metrics,including ATE and RPE to assess pose accuracy and rotational drift, respec-tively, alongside runtime, energy consumption, and resource usage to gaugetheir efficiency and practicality for real-world applications
Analytical study of the elastoplastic buckling of conical shells under external pressure
No full textInternational audienceAbstractPressure vessels are traditionally made up of cylindrical shells and hemispherical or ellipsoidal ends, but in some cases, conical sections are also present so as to ensure the transition between cylindrical sections of different radii. The buckling phenomenon is one of the main failure mode of such pressure equipments, due to the thinness of the components and the compressive stresses commonly undergone throughout standard loads like external pressure, and thus an essential dimensioning factor. If the buckling behavior of cylindrical and spherical shells has been widely investigated in the literature, the specific case of conical shells has received much less attention, all the more so in plasticity. Therefore, the present paper aims to address the problem of elastoplastic buckling of a conical shell under external pressure in an analytical way. This study is based on the plastic bifurcation theory and relies on the simplest possible hypotheses in terms of kinematics, constitutive law and boundary conditions. However, in absence of closed-form expressions, approximate solutions for the critical pressure are sought, based on the choice of appropriate shape functions in the framework of the Rayleigh–Ritz method. Unlike the case of cylindrical or spherical shells under external pressure which display uniform pre-critical stress states, the stress field appears to be heterogeneous in the length direction of a conical shell, so that three scenarios may occur. A conical shell may buckle elastically, entirely in the plastic range, or in an intermediate situation where the shell is partially elastic and plastic at the critical time. The present analytical solution is validated against reference numerical results obtained through finite element computations, considering a wide range of geometric and material parameters so as to cover all three scenarios
Softening effect on thick-walled tube inflation and bulging instability
No full textInternational audienceInflatable tubes often consist of rubber-like materials that exhibit stress softening over cyclic loading, referred to as the Mullins effect in the literature. A theoretical model is proposed to study the stress softening effect on the inflation of a thick-walled hyperelastic tube and the onset of an axisymmetric bulging. The three studied pre-loads (tension, torsion, inflation) create different softening distributions along the tube wall thickness. The resulting predicted bulging instability onset fluctuates, exhibiting a memory-like effect of the load type and intensity. A finite element analysis validates the theoretical analysis and evaluates the consequence of pre-softening loads on the post-bifurcation evolution. The theoretical analysis provides new elements for the design, control, and failure prevention of elastomer tube elements under inflation
Hardware Level Simulations of Fault Detection in RNS Accelerator
No full textInternational audienceProtection against fault attacks is required in some secure embedded systems. Several recent works deal with strong code protection at execution time. For symmetric cryptography, several fault detection solutions have been adapted from fault tolerance literature to protect the data-path and operations. For asymmetric cryptography based on modular arithmetic over large numbers (RSA, ECC, isogenies), fault detection remains a challenge. Using error correcting codes, triple modular redundancy, or duplication with comparison for large modular arithmetic is considered too costly.The Residue Number System (RNS) is a non-positional representation based on the Chinese Remainder Theorem, where an integer is represented by its residues modulo a set of pairwise coprime moduli. RNS offers a natural parallelism for addition and multiplication of large integers. RRNS uses an additional redundant modulus for fault detection and has been studied in signal processing implementations on small numbers. Theoretically, it allows effective fault detection with a overhead when using moduli. A few works used RRNS for fault detection in asymmetric cryptography, mainly RSA, but with few details regarding the hardware architecture and the fault model.We work on hardware accelerators for asymmetric cryptography primitives based on large numbers ( to bits) using RNS representations to benefit from its natural parallelism and fault detection capabilities. We study both arithmetic aspects (representations of numbers, algorithms) and low level architecture aspects. Up to now, there is no result on fault simulations at cycle accurate and bit accurate models in accelerator architectures. We try to complete state-of-art results with such experimental results to help us to select architecture details and parameters, and optimize our algorithms and arithmetic units using an accurate hardware model. We use the model of acomplete RNS accelerator with registers, functional units and internal memories. In this preliminary work, our accelerator model is complete but does not include pipeline registers to reduce the clock period (this is a future work). We performed intensive simulations for multiple faults models (mono vs multi-bit, stuck-at, bit-flip) in one or several registers of the accelerator. Our simulations confirm the high fault detection rate expected at theoretical level