Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Autotelic Reinforcement Learning in Multi-Agent Environments
No full textInternational audienceHow can a population of reinforcement learning agents autonomously learn a diversity of cooperative tasks in a shared environment? In the single-agent paradigm, goal-conditioned policies have been combined with intrinsic motivation mechanisms to endow agents with the ability to master a wide diversity of autonomously discovered goals. Transferring this idea to cooperative multi-agent systems (MAS) entails a challenge: intrinsically motivated agents that sample goals independently focus on a shared cooperative goal with low probability, impairing their learning performance. In this work, we propose a new learning paradigm for modeling such settings, the Decentralized Intrinsically Motivated Skill Acquisition Problem (Dec-IMSAP), and employ it to solve cooperative navigation tasks. Agents in a Dec-IMSAP are trained in a fully decentralized way, which comes in contrast to previous contributions in multi-goal MAS that consider a centralized goal-selection mechanism. Our empirical analysis indicates that a sufficient condition for efficiently learning a diversity of cooperative tasks is to ensure that a group aligns its goals, i.e., the agents pursue the same cooperative goal and learn to coordinate their actions through specialization. We introduce the Goal-coordination game, a fully-decentralized emergent communication algorithm, where goal alignment emerges from the maximization of individual rewards in multi-goal cooperative environments and show that it is able to reach equal performance to a centralized training baseline that guarantees aligned goals. To our knowledge, this is the first contribution addressing the problem of intrinsically motivated multi-agent goal exploration in a decentralized training paradigm
Scattering resonances in unbounded transmission problems with sign-changing coefficient
No full textInternational audienceIt is well-known that classical optical cavities can exhibit localized phenomena associated to scattering resonances, leading to numerical instabilities in approximating the solution. This result can be established via the ``quasimodes to resonances'' argument from the black-box scattering framework. Those localized phenomena concentrate at the inner boundary of the cavity and are called whispering gallery modes. In this paper we investigate scattering resonances for unbounded transmission problems with sign-changing coefficient (corresponding to optical cavities with negative optical properties, for example made of metamaterials). Due to the change of sign of optical properties, previous results cannot be applied directly, and interface phenomena at the metamaterial-dielectric interface (such as the so-called surface plasmons) emerge. We establish the existence of scattering resonances for arbitrary two-dimensional smooth metamaterial cavities. The proof relies on an asymptotic characterization of the resonances, and showing that problems with sign-changing coefficient naturally fit the black box scattering framework. Our asymptotic analysis reveals that, depending on the metamaterial's properties, scattering resonances situated closed to the real axis are associated to surface plasmons. Examples for several metamaterial cavities are provided
Evaluation des efforts d'origine inertielle dans les inclusions rigides
No full textInternational audienc
Parametric method for the assessment of fatigue damage for marine shaft lines
No full textInternational audienceAn original parametric method is introduced for the assessment of the fatigue life of marine shaft lines. The particularity of the method lies in the fact it introduces a relevant load modelling around a limited number of parameters specific to marine shaft lines while depicting the loading complexity (multiaxiality, mean stress effect, non-proportionality of the loading path). The method is designed to allow for assessments, in both in-use and pre-design phase shafts, towards a particular fatigue damage mode. The observations made in this study show two damage modes. On one hand, a damage mode issued from multiaxial cycles and associated with ship maneuvers, corresponding to HCF regime. On the other hand, there is a damage mode in the VHCF regime resulting from bending stress cycles due to the structure rotary bending
Modélisation de la combustion des matériaux énergétiques nouvelle génération
No full textUsed in civil and military rocket engines, composite propellants are characterized by a heterogeneous structure at the microscopic level. They consist mainly of oxidizer particles embedded in a polymer matrix known as the binder. The binder maintains the structural integrity of the propellant, and produces via its pyrolysis gases whose combustion is responsible for most of the heat release within the flame. The combustion products of the oxdizer particles oxidize the fuel generated by the binder.Ammonium perchlorate (AP) is a widely used oxidizing agent in composite propellants, mainly in association with a polymer binder such as hydroxytelechelic polybutadiene (HTPB). Next-generation propellants could include nitramines in their composition, such as hexogen (RDX), as a partial replacement for AP for certain specific purposes. In particular, reducing the amount of AP contained in the propellant can limit the formation of a visible trail at the engine exhaust, reducing the vehicle's signature.In this thesis, we propose to revisit the studies carried out on the combustion of conventional AP/HTPB propellants, and to characterize the effect of RDX inclusion in their composition.To this end, the first part of the thesis is dedicated to the development of combustion models for the ingredients under consideration: AP, homogenized AP/HTPB blend, and RDX.A chemical kinetics mechanism is assembled, capable of representing the chemical processes characteristic of the combustion of all these energetic materials. For each ingredient, a condensed-phase decomposition model is also formulated, to be associated with the kinetic mechanism. One-dimensional simulations are then carried out using a coupled flame/solid approach, in order to validate the whole on available experimental data.Due to their heterogeneous structure, the study of composite propellant combustion requires the use of multidimensional numerical methods. A second part of the thesis is therefore dedicated to the development and validation of a 2D calculation code, enabling the simulation of the combustion of an oxidizer particle surrounded by a layer of binder, in an axisymmetric configuration.In a final section, the combustion models and the computational code are used in conjunction to study the effect of RDX inclusion in a conventional AP/HTPB propellant. For the first time, the flame structure produced by such AP/HTPB/RDX propellant is obtained and characterized. The effect of pressure and RDX particle size on propellant regression is investigated, demonstrating the existence of different combustion regimes. Recommendations are made for optimizing the performance of this type of composition.Employés dans les moteurs fusée civiles et militaires, les propergols composites sont caractérisés par une structure hétérogène au niveau microscopique. Ils sont constitués principalement de particules oxydantes insérées au sein d’une matrice de polymère, appelée liant. Le liant maintient l’intégralité structurelle du propergol, et fournit par sa pyrolyse des gaz dont la combustion est responsable de la majorité du dégagement de chaleur dans la flamme. Les produits de combustion des particules oxydantes permettent l’oxydation du fuel généré par la pyrolyse du liant.Le perchlorate d'ammonium (PA) est un oxydant largement utilisé dans les propergols composites, principalement en association avec un liant polymère tel que le polybutadiène hydroxytéléchélique (PBHT). Les propergols nouvelle génération pourraient inclure des nitramines dans leur composition, telles que l'hexogène (RDX), en remplacement partiel du PA pour certaines fins spécifiques. En particulier, la réduction de la quantité de PA contenue dans le propergol permet de limiter la formation d'une traîné visible à l'échappement du moteur, réduisant la signature du véhicule.Dans cette thèse, nous proposons de revisiter les études menées sur la combustion des propergols PA/PBHT classiques, ainsi que de caractériser l'effet de l'inclusion du RDX dans leur composition.A cette fin, une première partie de la thèse est dédiée à la mise en place de modèles de combustion pour les ingrédients considérés: PA, mélange PA/PBHT homogénéisé, et RDX.Un mécanisme de cinétique chimique est mis au point, capable de représenter les processus chimiques caractéristiques de la combustion de l'ensemble de ces matériaux énergétique. Pour chaque ingrédient, un modèle de décomposition en phase condensée est par ailleurs formulé, pour être associé au mécanisme cinétique. Des simulations unidimensionnels sont alors réalisées en approche couplée flamme/solide, afin de valider l’ensemble sur les données expérimentales disponibles.De part leur structure hétérogène, l'étude de la combustion d'un propergol composite nécessite l'utilisation de méthodes numériques multidimensionnelles. Une seconde partie de la thèse est donc dédiée au développement et à la validation d’un code de calcul 2D, permettant la simulation de la combustion d’une particule oxydante entourée d'une couche de liant, en configuration axisymmétrique.Dans une dernière partie, les modèles de combustion mis au point sont utilisés conjointement avec le code de calcul développé afin d’étudier l’effet de l’inclusion du RDX dans un propergol PA/PBHT classique. Pour la première fois, la structure de la flamme produite par un propergol PA/PBHT/RDX est obtenue et caractérisée. Une étude est menée sur l'effet de la pression et de la taille des grains de RDX sur la régression du propergol, mettante en évidence l’existence de différents régimes de combustion. Des recommandations sont faites afin d'optimiser les performances de ce type de composition
Spectres de chargement pour le dimensionnement à la fatigue d’un véhicule automobile : identification, analyse et modélisation
No full textCar manufacturers are responsible for ensuring the reliability of all chassis components throughout the life of the vehicle. Nevertheless, no global standard framework exists. Therefore, they must rely on their own experience. For any design protocol, it is imperative to precisely define the loads seen by the parts. The load measures by the chassis system is even more complex to assess as it faces variable amplitude (depending on the life situations encountered) and multi-input load case (loads perceived at each wheel simultaneously). For fatigue design, the loads measured at the wheels are modelled as the sum of the ones induced by the manoeuvers, called Driven Road, and the ones induced by the road conditions, called Random Road. A validation process of the partition is illustrated using real time series. It benefits from the particularities of each loading type, time correlation and frequency characteristics, to consider the multi-input character of the load case. This model provides the manufacturer with additional elements to characterise the loadings.Les constructeurs automobiles se doivent d’assurer la fiabilité de l’ensemble des pièces de la liaison au sol sur toute la durée de vie du véhicule. Cependant, aucun cadre normatif global n’existe pour le dimensionnement de ces pièces. Ils doivent donc se baser sur leur expérience. Il est nécessaire pour tout protocole de dimensionnement de maîtriser précisément les chargements vus par les pièces. La sollicitation perçue par la liaison au sol est d’autant plus complexe à considérer qu’elle est à amplitude variable (suivant les situations de vie rencontrées) et multi-entrées (chargements perçus à chaque roue simultanément). Pour le dimensionnement à la fatigue, les sollicitations perçues aux roues sont modélisées comme la somme des chargements issus des manoeuvres, dits Driven Road, et de ceux induits par les aléas de la route, dits Random Road. Un processus de validation de cette partition est illustré à partir de mesures réelles. Il tire parti des particularités de chaque type de chargement, corrélation temporelle et caractéristiques fréquentielles, pour prendre en compte le caractère multi-entrées de la sollicitation. Ce modèle apporte au constructeur des éléments supplémentaires pour caractériser les chargements
Fatigue analysis of wind turbine composite blade using finite element method
No full textInternational audienceIn the past 30 years, wind turbine blades (WTB) have undergone significant development, increasing their size and introducing composites into manufacturing processes and using numerical simulation to assess their strength and structural integrity, helped increasing the number of installed wind turbine units as well as reducing the cost of wind generated energy. In this paper a DLoad subroutine was developed to assess monitor and evaluate the structural integrity of a large wind turbine blade under numerous static load scenarios. The fatigue study was carried using the finite element method, and the DLoad subroutine developed was used with ABAQUS finite Element analysis Software, and performed perfectly. The results show that the proposed layup parameters and the chosen composite materials gives to the wind turbine the desired structural strength. Furthermore, the DLoad subroutine for the fatigue study shows that the higher is the applied force the faster the blade fail. While, Hashin Criterion shows that the distribution of damage for the matrix and the fiber is all over the blade, but the failure only occurs after reaching an energy threshold which depends on the composite materials and the layup parameters used. Therefore, the chosen layup model will allow the wind turbine blade to withstand the extreme climatic conditions in the sea
TrustSoC: Light and Efficient Heterogeneous SoC Architecture, Secure-by-design
No full textInternational audienceIn recent years, heterogeneous SoCs, embedding multiple processor cores and programmable logic, have progressed in terms of complexity and performance. They embed more and more components of different natures. From a security point of view, this leads to an increase of the attack surface exploitable by an attacker. The goals of these attacks are to take control of the system and/or have access to sensitive data. To address this issue, in this article, we propose a novel heterogeneous SoC architecture called TrustSoC, which is secure-by-design. Our proposition presents an innovative way of partitioning the system into worlds to provide the designer with different levels of exclusion for the provision of security. Tiny and distributed hardware security wrappers apply policies and actively monitor the SoC communication bus to enforce these levels of security and prevent any unwanted behavior. TrustSoC is a novel proposition that considers both software and hardware approaches to secure the device. We demonstrate our approach by prototyping the security wrappers as well as their operating rules and show that TrustSoC requires minimal changes while significantly improving the state of the art on secure-by-design architectures
Symbolic Computation of Local Truncation Error for Approximate B-Series-Based Validated Simulation
No full textInternational audienc
COMPARISON OF BOUNDARY ELEMENT BASED AND PLANE WAVE APPROXIMATION COMPUTATIONS OF TARGET ECHO STRENGTHS
No full textInternational audienceIn naval defence applications, the knowledge of the Target echo strength (TES) of a submarine is of major interest, in order to optimize the scattered pressure that can be measured by an active sonar. In this contribution, we consider a rigid target and compute the TES using two methods: (i) the solution of the Helmholtz equation by reformulating it into a boundary integral equation with either a full space Green's function or a tailored Green's function, and (ii) the use of a plane wave approximation, well-suited for medium to high frequencies. In the first case, the use of a tailored Green's function adapted to the presence of a target reduces the cost of the numerical model. However, an integral equation still has to be solved. It is not the case with the plane wave approximation where the boundary pressure is not calculated but is considered proportional to the incoming wave. Numerical tests are performed to compare the efficiency and accuracy of each approach with respect to available numerical models developed on the submarine model "BeTSSi" -for Benchmark Target Strength Simulation -, under rigid hypothesis