Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Étude de l’acier inoxydable super-duplex 2507 fabriqué par procédé DED : de la fabricabilité à la caractérisation
No full textInternational audienc
A Route to 5,5-Dithiospiroketals and to Long Chain Monomers from the Biomass
No full textInternational audienceDithiospiroketals prepared by a double radical addition of a,a'-bis-xanthyl acetone to an alkene followed by ionic or thermal cleavage of the xanthate groups and acid catalyzed ring closure. A modification employing a-chloro-a'-xanthyl acetone unsymmetrical derivatives. Reductive removal of the sulfur atoms with Raney nickel furnishes long chain a,w-diacids, diols, and diamines. By using biosourced alkenes, monomers and polymers can be obtained where essentially all the carbons are derived from the biomass
Auralization of wind turbine noise using physics-based emission and propagation models
No full textInternational audienc
Domain invariant Q-learning for model-free robust continuous control under visual distractions
No full textInternational audienceEnd-to-end reinforcement learning on images showed significant performance progress in the recent years, especially with regularization to value estimation brought by data augmentation (Yarats et al., 2020). At the same time, domain randomization and representation learning helped push the limits of these algorithms in visually diverse environments, full of distractors and spurious noise, making RL more robust to unrelated visual features. We present DIQL, a method that combines risk invariant regularization and domain randomization to reduce out-of-distribution (OOD) generalization gap for temporal-difference learning. In this work, we draw a link by framing domain randomization as a richer extension of data augmentation to RL and support its generalized use. Our model-free approach improve baselines performances without the need of additional representation learning objectives and with limited additional computational cost. We show that DIQL outperforms existing methods on complex visuo-motor control environment with high visual perturbation. In particular, our approach achieves state-of the-art performance on the Distracting Control Suite benchmark, where we evaluate the robustness to a number of visual perturbators, as well as OOD generalization and extrapolation capabilities
A Mayer optimal control problem on Wasserstein spaces over Riemannian manifolds
No full textInternational audienc
Potential of the Noumea lagoon as a multi-mission cal/val site for past and future altimetry missions
No full textInternational audienceAt the interface between New-Caledonia and the Pacific Ocean, the Noumea Lagoon is an example of a dynamically complex coastal zone, that challenges the present and future altimetry mission. In this area, the SSH evolution is still an unresolved issue as sea level from altimetry and coastal tide gauge, and vertical land movements from GNSS, do not provide consistent information on long term trends (Aucan et al. 2017; Martínez-Asensio et al. 2019). This could be explained by multiple reasons, such as errors in the determination of GNSS Vertical Land Movements (VLM) due to mismodelled discontinuity (Ballu et al. 2019), a local elevation of the water body (setup) linked to waves and wind or the limitation of the altimetry data processing. We addressed this question through the GEOCEAN-NC 2019 oceanographic campaign. During 3 weeks, we deployed various instruments in the lagoon, with a particular interest for the intersection zone of 3 altimetry tracks (1 Jason track and 2 Sentinel-3a tracks). To obtain in-situ sea level variations over the entire altimeter period at this particular location, we virtually transfered the Noumea tide gauge using observations from a GNSS buoy (2 days) and a bottom pressure sensor (1 year). In the same time, we analysed the 20Hz along-track data from Jason and Sentinel-3a GDRs according to the standards used on other historical cal/val sites. For that, we determined the optimal GDRs correction parameters, and integrated geoid gradients from the XGM2019 model, validated thanks to the CalNaGeo GNSS carpet measurements. These two SSH time series (i.e. in-situ and altimetry) allow us to compute the altimeter biases over the entire Jason (1/2/3) and Sentinel-3a period. We show that the resulting altimeter biases are less than 10 cm for both Jason and Sentinel, with inter-mission biases consistent with other cal/val sites. With our 3 weeks field campaign, we have developed a relevant methodology to reanalyse 20 years of altimetry observations, and demonstrate the opportunities offer by the Noumea site to develop cal/val activities, for past, current and future missions. In this respect, several experiments are planned in 2023 to take advantage of the SWOT calibration phase 1-D repeat orbit
The effect of heat treatment on pseudoelastic behavior of spark plasma sintered NiTi
No full textInternational audienc
Numerical certification of Pareto optimality for biobjective nonlinear problems
No full textInternational audienceThe solution to a biobjective optimization problem is composed of a collection of trade-off solution called the Pareto set. Based on a computer assisted proof methodology, the present work studies the question of certifying numerically that a conjectured set is close to the Pareto set. Two situations are considered. First, we analyze the case where the conjectured set is directly provided: one objective is explicitly given as a function of the other. Second, we analyze the situation where the conjectured set is parameterized: both objectives are explicitly given as functions of a parameter. In both cases, we formulate the question of verifying that the conjectured set is close to the Pareto set as a global optimization problem. These situations are illustrated on a new class of extremal problems over convex polygons in the plane. The objectives are to maximize the area and perimeter of a polygon with a fixed diameter, for a given number of sides
Simulation numérique d'écoulements à surface libre avec la méthode des Eléments Finis
No full textInternational audienc
LoRaCog : a Protocol for Cognitive Radio-Based LoRa Network
No full textInternational audienceIn this paper, we propose a new protocol called LoRaCog to introduce the concept of Cognitive Radio (CR) in the LoRa network. LoRaCog will enable access to a wider spectrum than that of LoRaWAN by using the unutilized spectrum and thus has better efficiency without impacting the end devices’ battery consumption. LoRa networks are managed by LoRaWAN protocol and operate on the unlicensed Industrial, Scientific and Medical (ISM) band. LoRaWAN is one of thriving protocols for Low-Power Wide-Area Networks (LPWAN) implemented for the Internet of Things (IoT). With the growing demand for IoT, the unlicensed spectrum is expected to be congested, unlike the licensed spectrum, which is not fully utilized. This can be fairly balanced by applying CR to the LoRa network, where the End Devices (EDs) may change the operating channel opportunistically over the free/available licensed spectrum. Spectrum sensing, channel selection and channel availability relevance become essential features to be respected by the proposed protocol. The main objective of adding CR to LoRaWAN is reducing the congestion and maintaining LoRaWAN’s suitability for battery-operated devices. This is achieved by modifying LoRaWAN components such as the ED receive window RX2 rearrangement, spectrum sensing functionality by gateway (GW) for identifying unused channels, and reaching a decision on the unused channels by network server (NS). These changes will create LoRaCog meeting spectrum efficiency and maintain the same level of battery consumption as in LoRaWAN. Numerical simulations show a significant decrease in the rejected packet rate (more than 50%) with LoRaCog when more EDs use cognitive channels. As the results prove, LoRaWAN can reach above 50% rejected packets for the simulated environment versus 24% rejection for LoRaCog using only one additional channel (means total two channels). This means that the system can eliminate rejected packets almost completely when operating over the possible many channels. As well, these results show the flexibility in the system to utilize the available frequencies in an efficient and fair way. The results also reveal that a lower number of GWs is needed for LoRaCog from LoRaWAN to cover the same area