HAL: Hyper Article en Ligne
Not a member yet
3159010 research outputs found
Sort by
Beyond Paired Data: Self-Supervised UAV Geo-Localization from Reference Imagery Alone
No full textInternational audienceImage-based localization in GNSS-denied environments is critical for UAV autonomy. Existing state-of-the-art approaches rely on matching UAV images to geo-referenced satellite images; however, they typically require large-scale, paired UAV-satellite datasets for training. Such data are costly to acquire and often unavailable, limiting their applicability. To address this challenge, we adopt a training paradigm that removes the need for UAV imagery during training by learning directly from satellite-view reference images. This is achieved through a dedicated augmentation strategy that simulates the visual domain shift between satellite and real-world UAV views. We introduce CAEVL, an efficient model designed to exploit this paradigm, and validate it on ViLD, a new and challenging dataset of realworld UAV images that we release to the community. Our method achieves competitive performance compared to approaches trained with paired data, demonstrating its effectiveness and strong generalization capabilities.</div
Integrated air-rail scheduling: A branch-and-price approach for adaptive passenger-centric planning
No full textInternational audienc
Uncertainty in projected changes of Indian Summer Monsoon Rainfall by CMIP6 models
No full textInternational audienceA robust and trustworthy rainfall projection over the Indian landmass is vital for devising climate adaptation strategies. However, past studies show large inter-model spread in Indian Summer monsoon (ISM) rainfall projections thus calling for more detailed investigations on the underlying process. In the present study, we investigate this aspect using Coupled Model Intercomparison Project Phase 6 (CMIP6) model projections (Shared Socioeconomic Pathways, SSP5–8.5) and historical simulations. The Multi-Model Ensemble mean (MME) results show intensification of ISM rainfall at the end of the 21st century with ISM rainfall increasing by 1.6 ± 0.8 mm/day under SSP5–8.5 scenario. A moisture budget analysis for the MME further infers that the thermodynamic effect (TH) due to global warming plays a dominant role in enhancing ISM rainfall in the projections, with its dynamic counterpart (DY) assuming an additional contribution. It is also revealed that both DY and TH terms contribute to the inter-model uncertainty in ISM rainfall, but with DY dominating over the other this time. The inter-model uncertainty in DY and ISM rainfall changes is linked to inter-model spread in interhemispheric thermal contrast which in-turn depends on the diversity in Equilibrium Climate Sensitivity (ECS) and Global Mean Temperature (GMT) among the models. Intriguingly, when we remove the inter-model diversity in ECS through a GMT scaling, an Atlantic meridional surface temperature gradient, involving both land and ocean, emerges as a crucial driver in controlling the uncertainty in both DY and ISM rainfall changes, and drives large-scale monsoon circulation changes over African and the Indian subcontinents
"The sound is forc'd, the notes are few" : les Muses du romantisme anglais, tardives et fugitives ?
No full textInternational audienc
An adaptive sampling algorithm for data-generation to build a data-manifold for physical problem surrogate modeling
No full textInternational audiencePhysical models classically involved Partial Differential equations (PDE) and depending of their underlying complexity and the level of accuracy required, and known to be computationally expensive to numerically solve them. Thus, an idea would be to create a surrogate model relying on data generated by such solver. However, training such a model on an imbalanced data have been shown to be a very difficult task. Indeed, if the distribution of input leads to a poor response manifold representation, the model may not learn well and consequently, it may not predict the outcome with acceptable accuracy. In this work, we present an Adaptive Sampling Algorithm for Data Generation (ASADG) involving a physical model. As the initial input data may not accurately represent the response manifold in higher dimension, this algorithm iteratively adds input data into it. At each step the barycenter of each simplicial complex, that the manifold is discretized into, is added as new input data, if a certain threshold is satisfied. We demonstrate the efficiency of the data sampling algorithm in comparison with LHS method for generating more representative input data. To do so, we focus on the construction of a harmonic transport problem metamodel by generating data through a classical solver. By using such algorithm, it is possible to generate the same number of input data as LHS while providing a better representation of the response manifold
Numerical investigation of the effect of defect population on the fatigue strength anisotropy of alloys fabricated by L-PBF
No full textInternational audienceThis study investigates the impact of defects on the fatigue strength anisotropy of Ti-6Al-4V alloy fabricated by Laser Powder Bed Fusion (L-PBF). Finite element simulations are performed to analyze two defect populations, lack of fusion (LoF) defects and gas pores, under multiaxial proportional loading conditions. Results demonstrate that fatigue strength is highly sensitive to the polar angle representing the defect orientation with respect to the mechanical loading and is primarily influenced by defect morphology, transitioning from low sphericity (LoF defects) to high sphericity (gas pores). The anisotropic behavior is particularly pronounced under tensile loading, where the morphology of LoF defects and loading directions exhibit a notable influence. Shear loading exhibits reduced sensitivity to these factors. The proposed statistical methodology allow to determine the fatigue strength anisotropy for any multiaxial loading configuration and defect population. The numerically obtained fatigue strength anisotropy matches the one obtained experimentally in the literature and non-local analysis smoothens the fatigue strength surface response, leading to a reduction in the anisotropic factor ()
Sensitivity of thermal evapotranspiration models to surface and atmospheric drivers across ecosystems and aridity
No full textInternational audienceEvapotranspiration (ET) lies at the core of the energy-water-carbon coupling, particularly under changing climate conditions. Yet, the sensitivity of ET models to key environmental drivers remains insufficiently understood, especially in understanding how thermal-based ET models respond to distinct influences of soil and atmospheric water stress across ecosystems. In this study, we examine the sensitivity of ET to key environmental drivers, including land surface temperature (LST), air temperature (TA), vapor pressure deficit (VPD), downward solar radiation (DSR), and fractional vegetation cover (FVC), using three representative thermal remote sensing (RS) models (STIC, TSEB, and SPARSE) together with global eddy covariance measurements.At the global scale, variance-based sensitivity analysis (Sobol' method) reveals a transition in the dominant driver of ET sensitivity from water-limited to energy-limited regimes: soil dryness (indicated by LST − TA) dominates ET variability up to an aridity index (ratio of precipitation to reference ET) of 0.54 (± 0.06), beyond which DSR becomes the primary driver. Seasonal variability and ET partitioning emphasize the critical role of soil dryness in driving soil evaporation variability, particularly during growing seasons. Furthermore, a water stress test is conducted across four representative sites with varying vegetation cover types. Results show that ET sensitivity to soil dryness nearly doubles during the drought period compared to climatological norms at the grassland site. In contrast, transpiration in forests is more strongly influenced by VPD under moderate drought stress. Analysis indicates that soil dryness generally exerts stronger control on ET than VPD. However, when vegetation cover exceeds 0.49, the influence of VPD anomalies on ET becomes comparable to soil dryness stress.This research advances our understanding of ET dynamics under increasing drought frequency and intensity. It highlights the potential of forthcoming high-resolution thermal-based RS ET products for early drought hazard warnings, climate-resilient decision-making, and sustainable agricultural water management
Experimental and numerical investigation of the impact force generated by cylindrical ice water pellets
No full textInternational audienceThis study focuses on the force-time response of cylindrical water ice specimens subjected to impact loadings. Spherical specimens are traditionally used to characterize the impact behavior of water ice. However, they cannot be used to study the geometric effects induced by a cylindrical shape. Impact tests were carried out on a Hopkinson bar at 30 m.s -1 . These tests have demonstrated the importance of the impact angle in terms of both the increase in the load and the peak force at impact. Contrarily to what was observed for tensile spalling test, porosity has no noticeable impact on the maximum peak force measured here. The importance of the impact angle is illustrated by comparing the mechanical response of ice spheres with pellet cylinders for equivalent kinetic energies and temperatures
Certified Enumeration of AI Explanations: A Focus on Monotonic Classifiers
No full textInternational audienceThe theory of minimal explanations offers a rigorous, model-based solution to the problem of producing explanations for the decisions of AI models. In some high-stakes contexts, there is a need to generate all possible explanations for a particular decision using certified programs, whose output can be trusted. We used the proof assistant Coq to certify a recently proposed algorithm for the enumeration of explanations in the case of monotonic classifiers. Our experimental results on the extracted code showcase the scalability of this approach, underscoring its potential for improving trust and reliability in AI systems