HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Facial Empathy Analysis Through Deep Learning and Computer Vision Techniques in Mixed Reality Environments
No full textInternational audienceThis paper introduces a novel approach for facial empathy analysis using deep learning and computer vision techniques within mixed reality environments. The primary objective is to detect and quantify empathic responses based on facial expressions, establishing the link between empathy and facial expressions. We propose the Deep Convolutional Neural Network with the Exponential Linear Unit activation function (ELU-DCNN). We moreover design an augmented reality platform with two main features (i). virtual overlay of a VR headset on the user’s face and (ii). facial emotion recognition for users wearing the VR headset. Our target is to analyse facial expressions in immersed environments in order to assess the empathy of users while being immersed in specific environments. Our results analyse the feasibility and effectiveness of these models in detecting and quantifying empathy through facial expressions. This work contributes to the growing field of affective computing and highlights th e potential of integrating advanced computer vision techniques in mixed reality applications to better understand human emotional responses
Actes du SDFIA 2025: Premier symposium doctoral francophone en intelligence artificielle
No full textInternational audienceLes actes du SDFIA 2025 rassemblent des travaux doctoraux francophones couvrant un large spectre de l’IA, de la vision par ordinateur aux systèmes multi‑agents, de l’IA explicable aux modèles robustes d’estimation, jusqu’aux applications sécurité réseau et énergie. Organisé le 9 octobre 2025 à l’issue de l’école d’automne RobIA’25, l’événement met l’accent sur la reproductibilité, la diffusion open source et la rigueur méthodologique. Les articles longs présentent des contributions expérimentales et comparatives (classification d’images, génération pour classes rares, détection d’intrusions), tandis que les articles courts explorent consensus du second ordre, ViT pour la détection de piétons, SSL pour obstacles routiers, et filtres de Kalman robustes. L’ensemble témoigne de la vitalité de la recherche doctorale francophone, avec des résultats applicables à l’agriculture, la mobilité, la cybersécurité, l’éducation et le photovoltaïque
A Multi-Start Tabu Search with Set Partitioning for the Green VRP
No full textInternational audienceThis paper tackles the Green Vehicle Routing Problem (GVRP), where vehicles with limited driving range must visit customers while recharging at Alternative Fuel Stations (AFSs). We propose a Multi-Start Tabu Search with Set Partitioning (MSTS-SP) approach structured in two phases. In the first phase, MSTS-SP uses a new constructive heuristic, Randomized Sectoring with Repair, to generate diverse initial solutions, which are then improved through multiple independent tabu search runs. The high-quality routes found during these runs are collected into a global pool. In the second phase, an exact set partitioning model is applied to this pool to select the best combination of routes. Computational experiments on 52 GVRP benchmark instances show that MSTS-SP matches 46 known best solutions (88%) and improves upon the best known solution for one large instance. These results demonstrate that MSTS-SP offers a competitive balance between solution quality and computational efficiency compared to state-of-the-art methods
Effect of variations in parameters on the crystallization of mordenite zeolite
No full textInternational audienceIntroduction/purpose: This study highlights the importance of synthesis parameters in the crystallization of mordenite zeolite. Precise control of the SiO₂/Al₂O₃ ratio, alkalinity and crystallization time results in mordenite crystals with high crystallinity and optimum purity, while avoiding the formation of secondary or amorphous phases. Methods: The materials were prepared by the hydrothermal method, using silica gel and sodium aluminate as sources of silicon and aluminum, respectively. Several synthesis parameters were varied, including SiO₂/Al₂O₃ molar ratio, alkalinity (OH-/Si), as well as crystallization time, in order to assess their effect on mordenite crystal formation. Experiments were carried out at a constant temperature of 170°C and 190°C. Results: The results show that the SiO₂/Al₂O₃ ratio plays a crucial role in crystal formation. A low ratio, such as 15, combined with high alkalinity, favors the formation of analcime crystals. On the other hand, a high ratio, such as 30, leads to the formation of mordenite crystals with high crystallinity and purity. However, when this ratio reaches 60 and is combined with low alkalinity, crystalline nucleation is impeded, leading to the formation of an amorphous material. Concerning alkalinity (OH-/Si), the values of 0.39 and 0.49 result in pure, well-crystallized mordenite crystals, while higher values, such as 0.59, lead to the formation of secondary phases. With regard to crystallization time, the periods of 48 and 72 hours at 170°C produced pure, well-crystallized mordenite crystals. Conclusion: This study highlights the importance of synthesis parameters in the crystallization of mordenite zeolite. Precise control of the SiO₂/Al₂O₃ ratio, alkalinity and crystallization time results in mordenite crystals with high crystallinity and optimum purity, while avoiding the formation of secondary or amorphous phases
Predictive Road Optimization Strategy for Efficient Message Delivery in Delay-Tolerant Networks
No full textInternational audienc
Convergence acceleration of a nonlinear solver using statistical learning
No full textInternational audienceNumerical simulation is widely used in industrial product design. However, its application in engineering often involves significant computational costs and expertise.This research project, carried out in collaboration with Michelin, ENS Paris-Saclay, and UTC, aims to investigate the contribution of learning-based methods (possibly physics-informed) to accelerate the convergence of a nonlinear solver. In particular, we focus on predicting the initialization of Newton-type solution algorithms by leveraging dimensionality reduction techniques and digital twin learning models. The effectiveness of these approaches is assessed on industrially relevant application cases related to tire modeling, where solving large-scale nonlinear systems remains a major computational challenge
An Octree-based adaptive moving thermal-fluid framework for efficient multi-scale and multi-physics simulation of laser-based manufacturing processes
No full textInternational audienceMultiphysics and multi-scale thermal-fluid models are essential for understanding the coupling of physical phenomena, enabling the optimization of welding and additive manufacturing process parameters. However, the thermal-fluid simulations are extremely time-consuming due to the need for a fine mesh (approximately 100µm) and small time steps (around 10 -5 s), which restrict the modeling domain to very small dimension. This paper proposes a multiphysics, multi-scale thermal-fluid model based on the Moving Thermal-Fluid (MTF) framework combined with Octree-based Adaptative Mesh Refinement (AMR) that can solve heat and mass transfer problem in an optimal way. The MTF framework consists of solving the thermal-fluid problem only within a small, moving region containing the melt pool, while calculating a heat transfer problem in the rest of the domain. Therefore, much fewer degrees of freedom (DOF) should be solved for a gvien mesh.The Octree-based AMR will create a fine mesh in the moving region containing the melt pool and a coarser mesh in the rest of the region. Thanks to the presence of Octree structure, the remeshing and variables transfer for the nodes and Gauss point can be performed in an analytical way with negligible CPU time. To validate the proposed approach, simulations of a laser welding benchmark and a single-track direct energy deposition process were conducted using both the Octree-based MTF framework and the MTF</p
Sustainable 2-Phenylethanol Production: Co-Cultivation of Yarrowia lipolytica Strains in Mixed Agro-Industrial By-Products
No full textInternational audienceThe bioproduction of 2-phenylethanol (2-PE), a high-value aromatic compound widely used in the fragrance, cosmetic, food and beverage, and pharmaceutical industries, through yeast fermentation offers a sustainable alternative to chemical synthesis and rose extraction. This study explores the fermentation of Yarrowia lipolytica strains using mixed agro-industrial by-products as substrates to produce 2-PE via de novo synthesis, without supplementation with the costly precursor L-phenylalanine. Y. lipolytica strains were genetically engineered to enhance flux through the shikimate pathway and enable the hydrolysis of a broader range of substrates. The culture media consisted solely of a mixture of agro-industrial by-products: sugar beet molasses (SBM), brewer’s spent grain (BSG) pressing extract, and chicory root (CR) pressing extract, serving as the primary carbon and nitrogen sources without the addition of nutrients, minerals, synthetic, complex ingredients, or costly additives. The co-culture approach enhanced substrate utilization, leading to an increase in 2-PE titers, reaching approximately 2.5 g/L 2-PE production after 240 h of fermentation. This study demonstrates the feasibility of integrating co-culture fermentation and agro-industrial waste valorization for sustainable 2-PE production, offering a scalable bioprocess for industrial applications
Blood flow visualization and quantification in the carotid vascular tree by phase contrast MRI
No full textInternational audiencePurpose:The objective is to build a phase-contrast (PC) MRI protocol, consistent with clinical practice, to provide a 3D blood flow visualization and quantification of hemodynamic parameters in the complete carotid vascular tree. Methods:The protocol composed of 2D and 4D PC-MRI sequences was applied on 6 volunteers and then on one patient diagnosed with facial cancer to prove the feasibility of clinical translation. The vessel geometry was reconstructed from the 4D sequences and the hemodynamic parameters quantified in the common, internal and external carotids and in the facial artery. Wall shear stresses (WSS) were quantified from the 2D PC-MRI sequences to benefit from their higher resolution.Results: Time evolution of the three-dimensional blood flow velocity and vorticity fields was successfully obtained in all the branches of the carotid vascular tree despite the large range of sizes.Consistent maps of blood flow distribution were provided by normalizing the local blood flows by that in the common carotid artery. They indicated that 72.4% (± 3.9 %) of blood flows into in the internal carotid. WSS is higher in the internal (0.95 Pa at peak systole) than in the external carotid (0.53 Pa) and facial artery (0.15 Pa). Conclusion:A PC-MRI protocol, applicable on patients, was designed to quantify hemodynamic parameters in vessels ranging from a few millimeters to the centimeter in diameter. It provided a complete characterization of the hemodynamic condition evolution along the carotid vascular tree, and reference values to be compared to in case of pathology.</div
Structural and electrical properties of alkyl/perfluoroalkyl-thiol self-assembled monolayer on germanium as passivation layer
No full textFor several decades, germanium has been considered one of most attractive semiconductors in various fields, in particular within the channel design of the next generation metal-oxide-semiconductor field effect transistors (MOSFET). However, the use of germanium iscompromised by the poor quality of germanium oxide. This motivates the present study,focused on the structural and electrical characterization of self-assembled monolayers of fluoro2alkylthiols and variable length alkylthiols on Ge, which may be considered alternate passivationand insulating layers to replace germanium oxide. The de-oxidation/grafting technique inethanol-water mixed solution has been adapted in this work. The results show that it providesbetter results than the more usual acid treatment. Indeed, such a method has allowed us to obtainsmoother functionalized Ge surfaces with well-organized self-assembled monolayers (SAMs),assessed by ellipsometry, goniometry, and atomic force microscopy (AFM). X-rayphotoelectron spectroscopy (XPS) analyses demonstrate the removal of oxide from as-functionalized Ge surface. From current-voltage measurements of the various SAMs using liquid gallium-indium eutectic contacts, it appears that fluoro-alkyl SAMs have allowed to decrease the current by more than two decades compared to bare (oxidized) Ge surfaces.Statistical analyses of the electrical characteristics are correlated with spectroscopic studies of the molecular levels, using inverse photoemission spectroscopy (IPES) for probing the unoccupied levels (LUMO), XPS for the occupied levels (HOMO), and DFT calculation