Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Computation of a fluid-structure Green's function using a BEM-BEM coupling
No full textInternational audienceIn order to determine the elasto-acoustic noise produced by a boat hull excited by a turbulent boundary layer, we propose a numerical method to compute the acoustic scattering by an elastic body surrounded by a fluid. To reduce the computational costs a Boundary Element Method (BEM) is used. Since the turbulent flow along the hull is known only statistically, a formulation combining the free field acoustic and elastic Green's functions is not adequate. A better suited choice is to determine a global Green's function satisfying the transmission conditions of the fluid-structure problem. The boundary integral representation of the scattered pressure is then simplified. This so-called tailored Green's function is determined by solving an acoustic/elastic coupled problem with a BEM. Here we focus on a particular difficulty: when the source is close to the surface, the numerical accuracy of the Green's function deteriorates. We describe a method to regularize our BEM scheme in this context. We validate the method for the problem of an elastic sphere in water
SAC-GLAM: Improving Online RL for LLM agents with Soft Actor-Critic and Hindsight Relabeling
No full textInternational audienceThe past years have seen Large Language Models (LLMs) strive not only as generative models but also as agents solving textual sequential decision-making tasks. When facing complex environments where their zero-shot abilities are insufficient, recent work showed online Reinforcement Learning (RL) could be used for the LLM agent to discover and learn efficient strategies interactively. However, most prior work sticks to on-policy algorithms, which greatly reduces the scope of methods such agents could use for both exploration and exploitation, such as experience replay and hindsight relabeling. Yet, such methods may be key for LLM learning agents, and in particular when designing autonomous intrinsically motivated agents sampling and pursuing their own goals (i.e. autotelic agents). This paper presents and studies an adaptation of Soft Actor-Critic and hindsight relabeling to LLM agents. Our method not only paves the path towards autotelic LLM agents that learn online but can also outperform on-policy methods in more classic multi-goal RL environments
Effects of pulsed laser and electron-beam irradiation on protected aeronautical CFRP composite laminate to enhance the modelling of lightning-strike damage
No full textInternational audienceIn this paper, damage produced by lightning strike, laser shock and electron beam deposition on a protected Carbon Fiber Reinforced Plastic composite laminate is studied in order to find analogies of effects and damage between these experimental means. As lightning strike physics on CFRP coated with a Lightning Strike Protection and paint is not fully understood, these analogies could be able to enhance lightning strike modelling by potentially uncoupling the physics at hand and having access to additional measurement instruments. The different experimental setups are briefly described before analyzing the damage response of the aeronautical CFRP protected using an Expanded Copper Foil and coated in aeronautical paint. Eventually the results are compared to build potential analogies able to enhance lightning strike modelling
Adapting Gaussian Mixture Model Training to Embedded/Edge Devices: A Low I/O, Deadline-Aware and Energy Efficient Design
No full textInternational audienceThis work focuses on the Gaussian Mixture Model (GMM), a machine learning model used for density estimation and cluster analysis in healthcare, networking, etc. The Expectation-Maximization (EM) algorithm is commonly used to train GMMs. One of the main challenges facing this algorithm when running on embedded systems is the crippling memory constraints. In fact, EM requires several scans of the dataset and we observed that when the dataset cannot fully reside in the main memory, its execution is dramatically slowed down by I/O movements. In this paper, we present an optimization of the EM algorithm for GMMs that reduces the number of I/O operations thanks to two main contributions: (1) A divide-and-conquer strategy that divides the dataset into chunks, learns the GMM separately on each chunk and combines the results incrementally. By doing so, we prevent data from being swapped several times during the learning process. (2) Restricting the training on a subset of data whose volume is inferred online using data properties while producing good accuracy. On average, our results show a 63% improvement in overall execution time with comparable accuracy. We also adapted GMM learning to run in a limited time budget while hitting a good trade-off between execution time and energy consumption. This solution succeeded in meeting the fixed deadline in 100% of the cases and in reducing the energy consumption by up to 68.77%
A Virtual Reference Point Kinematic Guidance Law for 3-D Path-Following of Autonomous Underwater Vehicles
No full textInternational audienceThis work presents a novel method for 3-D path-following and path-tracking of Autonomous Underwater Vehicles (AUVs) using the concept of a Virtual Reference Point (VRP) and a kinematic guidance principle. The origins of the along-, cross-and vertical-track errors are proven globally exponentially stable (GES) using Lyapunov stability analysis. The kinematic guidance law exploits the design flexibility of a user-defined VRP in conjunction with a feedback linearizing controller. In addition, a novel concept called the Handy Matrix is introduced and applied to shape the kinematic equations such that the AUV's nonactuated degrees of freedom (DOFs) can be controlled in a 3-D path-following scenario. The case study considers the Remus 100, a torpedo-shaped underactuated AUV, performing a 3-D path-following maneuver. The computer simulations show that the kinematic guidance law shows excellent tracking performance and stability even in the presence of ocean currents and white measurement noise
Extending the Discrete Element Method to Account for Dynamic Confinement and Strain-Rate Effects for Simulating Hard Impacts on Concrete Targets
No full textInternational audienceConcrete plays a pivotal role as a foundational material in critical infrastructure, particularly in nuclear plants. Given the imperative for robustness and safety in such contexts, the design of concrete structures necessitates methodologies capable of precisely predicting damage resulting from impacts. When subjected to impact, concrete experiences high loading rates and significant triaxial stresses in the vicinity of the impacting object, potentially resulting in fragmentation, pore closure and projectile penetration. In addressing these challenges, the discrete element method (DEM) emerges as a suitable approach primarily due to its inherent ability to model discontinuities such as cracking and fragmenta-tion. Within this framework, DEM employing spherical discrete elements (DE) has been implemented into Europlexus, a fast transient dynamics finite element (FE) code. This paper presents a refined constitutive DEM model for concrete, especially accounting for porosity closure under high confine-ment and the effect of strain-rate on tensile strength and fracture energy. The calibration of constitutive parameters is conducted through the simulations of a series of quasi-static (QS) tests encompassing tension, compression and triaxial compression. The strain-rate dependency parameters are identified through dynamic tensile tests conducted using a split-Hopkinson pressure bar apparatus. The whole constitutive model is presented as well as its calibration. Finally, the validation of the DEM approach is demonstrated through simulations of penetration and perforation tests conducted on concrete targets
Application of a generic path-following method to phase-field fracture
No full textInternational audienceThe phase-field approach to fracture has emerged as a powerful tool to simulate the nucleation and growth of cracks in a structure. In the past two decades, it has been extensively applied to fracture problems as it captures crack initiation, propagation, and interaction without explicitly tracking the crack path. One of the most popular algorithms to solve phase-field problems is alternate minimization. However, it can suffer from slow convergence, especially when dealing with unstable crack propagation. Moreover, force-controlled loading often leads to unstable crack propagation and the lack of equilibrium solution after the crack propagation, preventing their use. Path-following methods offer a promising solution to those limitations, enabling the tracking of unstable crack propagation while preserving the equilibrium during the whole loading (Rastiello et al., 2022). Based on various control strategies, these methods also improve the solver stability. Singh et al. (2016) and May et al. (2016) proposed path-following approaches specifically tailored to the resolution scheme of Miehe et al. (2010) based on crack surface and fracture dissipation. Additionally, Wu (2018) adapted the nodal displacement control (Borst, 1987) and the fracture surface control (Singh et al., 2016) to the alternate minimization. Nevertheless, the first approach is problem-dependent, and the second approach may fail under force loading (Rastiello et al., 2022). This work proposes a generic path-following method applicable to various fracture problems, regardless of geometry, boundary conditions, or fracture model complexity, by leveraging the maximum strain increment control (Chen & Schreyer, 1990). This method is model-independent, as it relies solely on the displacement field, and problem-independent, it does not rely on a specific choice of control DOF. After presenting the modified alternate minimization solver, we demonstrate its effectiveness through simulations of crack propagation in the SENT test. The results are compared to a semi-analytical solution based on LEFM and to the alternate minimization solution. Notably, the classic alternate minimization fails to capture the snap-back (instability under displacement control) observed in the semi-analytical method. The proposed approach correctly capturesthis phenomenon, which converges towards the semi-analytical solution. Then, this method is also applied to the simulation of Compact Tension (CT) experiments, in which the selectionof numerical boundary conditions at the pinhole significantly influences the fracture behavior (Triclot et al., 2023). The proposed solver renders the application of force boundary conditions possible, better representing the experimental conditions.ReferencesBorst, R. de. (1987). Computation of post-bifurcation and post-failure behavior of strain-softening solids. Computers & Structures, 25(2), 211–224. https://doi.org/10.1016/0045-7949(87)90144-1Chen, Z., & Schreyer, H. L. (1990). A numerical solution scheme for softening problems involving total strain control. Computers & Structures, 37 (6), 1043–1050. https://doi.org/10.1016/0045-7949(90)90016-UMay, S., Vignollet, J., & Borst, R. de. (2016). A new arc-length control method based on the rates of the internal and the dissipated energy. Engineering Computations, 33(1), 100–115. https://doi.org/10.1108/EC-02-2015-0044Miehe, C., Hofacker, M., & Welschinger, F. (2010). A phase field model for rate-independent crack propagation: Robust algorithmic implementation based on operator splits. Computer Methods in Applied Mechanics and Engineering, 199(45), 2765–2778. https://doi.org/10.1016/j.cma.2010.04.011Rastiello, G., Oliveira, H. L., & Millard, A. (2022). Path-following methods for unstable structural responses induced by strain softening: A critical review. Comptes Rendus. Mécanique, 350, 205–236. https://doi.org/10.5802/crmeca.112Singh, N., Verhoosel, C. V., Borst, R. de, & Brummelen, E. H. van. (2016). A fracture-controlled path-following technique for phase-field modeling of brittle fracture. Finite Elements in Analysis and Design, 113, 14–29. https://doi.org/10.1016/j.finel.2015.12.005Triclot, J., Corre, T., Gravouil, A., & Lazarus, V. (2023). Key role of boundary conditions for the 2D modeling of crack propagation in linear elastic compact tension tests. Engineering Fracture Mechanics, 277, 109012. https://doi.org/10.1016/j.engfracmech.2022.109012Wu, J.-Y. (2018). Robust numerical implementation of non-standard phase-field damage models for failure in solids. Computer Methods in Applied Mechanics and Engineering, 340, 767–797. https://doi.org/10.1016/j.cma.2018.06.00
Palladium-Catalyzed C8-Fluoroalkylation and C8-Fluoroalkenylation of Naphthalenes by C-H Activation
No full textInternational audienceIn this study, two novel C8‐functionalization methods of 1‐carbonylnaphthalenes were developed. Leveraging iodonium triflates as coupling partners, we disclose palladium‐catalyzed C8‐fluoroalkylation and C8‐fluoroalkenylation processes. These protocols exhibit broad applicability across diverse naphthalene derivatives, owing to carbonyl directing groups at the C1 position which can be retained for subsequent applications or selectively removed, enhancing the versatility of the developed methodologies
Feature Expansion and enhanced Compression for Class Incremental Learning
No full textClass incremental learning consists in training discriminative models to classify an increasing number of classes over time. However, doing so using only the newly added class data leads to the known problem of catastrophic forgetting of the previous classes. Recently, dynamic deep learning architectures have been shown to exhibit a better stability-plasticity trade-off by dynamically adding new feature extractors to the model in order to learn new classes followed by a compression step to scale the model back to its original size, thus avoiding a growing number of parameters. In this context, we propose a new algorithm that enhances the compression of previous class knowledge by cutting and mixing patches of previous class samples with the new images during compression using our Rehearsal-CutMix method. We show that this new data augmentation reduces catastrophic forgetting by specifically targeting past class information and improving its compression. Extensive experiments performed on the CIFAR and ImageNet datasets under diverse incremental learning evaluation protocols demonstrate that our approach consistently outperforms the state-of-the-art . The code will be made available upon publication of our work
Scattering Operators and High-Order Statistics along with Elastography to Identify and Characterize Salivary Gland Abnormalities
No full textInternational audienceThis study is the result of a collaborative project between an engineering school Ecole Nationale Supérieure de Techniques Avancées Bretagne (ENSTA Bretagne, Lab-STICC UMR CNRS 6285) and a Medical Research Group “Groupe d’Etude de Thrombose de Bretagne Occidentale” (EA 3878 GETBO) associated with the INvestigation Network On Venous Thrombo-Embolism (INNOVTE) and the clinical investigation center (CIC Inserm 1408) of the Brest University Hospital (CHU - Brest). Our study considers the abnormalities of salivary glands. It focuses on the detection and characterization of the syndrome of Gougerot-Sjögren. To reach our goals, we are collecting ultrasonography (echogenicity) and elastography (stiffness) images of the salivary glands of several patients. This paper resumes our proposed preprocessing steps and approaches used to create and preprocess our database using the Canon Aplio system. Our approaches to characterize the salivary glands with ultrasound images are also described. We propose to extract features from these kinds of ultrasound images using two approaches: one based on wavelets (the scattering operators) and another one based on high-order statistics (multicorrelations). Then, the obtained features are analyzed using several classification technics (principal component analysis, k-means, and spectral clustering). Experimental results are presented and discussed. We should highlight the fact that our approach can reach an outstanding result that can be compared to experts’ outcomes