HAL Arts et Métiers
Not a member yet
14127 research outputs found
Sort by
Reconfigurable flexible haptic interface using localized friction modulation
No full textInternational audienceThis paper describes the development, the characterizationand the evaluation of a flexible haptic interface usinglocalized friction reduction. Based on previous work, the surfacedeveloped is composed of several haptic resonators vibratingat an ultrasonic frequency, driven by piezoelectric actuators,and associated with a polymer matrix. The solution combinesthe advantages of a rigid haptic surface implementing frictionreduction and the conformability of a 75 μm thick polymersheet. The development and the realization of the prototypeare presented in this paper. By powering or not poweringthe actuators, it is possible to display simple tactile shapes.Tribological measurements confirm that the friction reductionproduced by each haptic resonator follows the desired shape.Finally, two studies were carried out to show the usefulness ofthe device in two typical use cases. In the first one, participantscould recognize shapes, with an average success rate of 96 % anda detection time around 14 s. In the second one, two participantsexplore the device, each with different haptic feedback. The studyshowed that they could discriminate the tactile pattern presentedto them with a success rate of 89 %
Converging narrow-channel flow of a super-critical fluid
No full textInternational audienceThe solution of a supercritical fluid flowing into a constricted narrow channel is presented in this study. The compressible Navier-Stokes equations in the lubrication limit coupled with the energy equation and the isothermal and non-isothermal van der Waals fluid and perfect gas have been solved. In order to find the semi-analytical solution of these non-linear coupled equations, homogenization technique in the transverse direction has been applied. Because of the high compressibility and high thermal expansion of supercritical fluids, waviness is observed in the flow and thermal fields near the exit of the channel. This effect is attributed to the channel constriction where the slope is maximum, where a strong coupling between the pressure and density gradients exists. Moreover, the density difference between the exit and inlet of the channel drastically increases when one approaches the critical point, corroborating the data from existing literature
Reduced-order modeling for nonlinear vibrations of structures
No full textInternational audienceThis chapter is devoted to the presentation of model-order reduction techniques that are used in the field of structural vibrations. A special emphasis is placed on substructuring methods for localized nonlinearities and on nonlinear normal modes defined via invariant manifolds for distributed smooth nonlinearities, as key tools to perform efficient yet accurate dimensional reductions. Other reduction techniques such as proper orthogonal decomposition, implicit condensation, and modal derivatives are also briefly covered at the end of the survey. The contents of this chapter were written for the Handbook of Nonlinear Dynamics during the Summer of 2024
Machine learning (ML) based reduced order modelling (ROM) for linear and non-linear solid and structural mechanics
No full textInternational audienceMultiple model reduction techniques have been proposed to tackle linear and non-linear problems. Intrusive model order reduction techniques exhibit high accuracy levels; however, they are rarely used as a standalone industrial tool because of the high level of expertise required in the construction and usage of these techniques. Moreover, the computation time benefit is compromised for highly non-linear problems. On the other hand, non-intrusive methods often struggle with accuracy in non-linear cases, typically requiring a large design of experiments and a large number of snapshots in order to achieve reliable performance. However, generating the stiffness matrix in a non-intrusive approach presents an optimal way to align accuracy with efficiency, combining the advantages of both intrusive and non-intrusive methods. This work introduces a minimally intrusive model order reduction technique that employs machine learning within a Proper Orthogonal Decomposition framework to achieve this alliance. By leveraging outputs from commercial full-order models, this method constructs a reduced-order model that operates effectively without requiring expert user intervention. The proposed technique is capable of approximating linear non-affine as well as non-linear terms. It is showcased for linear and non-linear structural mechanics problems.</div
Effect of shot-peening compaction process on the microstructure and corrosion behavior of an Al multiparticulate coating
No full textInternational audienceAbstract To enhance the corrosion protection of a steel substrate coated with aluminum powder embedded in a sol–gel matrix, a shot-peening compaction process of the coating with sodium bicarbonate beads was developed. The initial coating exhibited a discontinuous microstructure characterized by large cracks and porosities between the Al particles throughout its depth. Microstructural, mechanical and corrosion properties were evaluated before and after shot-peening compaction. Depending on process parameters (impact velocity and coverage rate), the treatment enabled total or partial closure of the cracks and porosities, resulting in a relative density increase of up to 28%, as quantified by image analysis. A gradient in microstructure and mechanical properties was observed across the coating depth, with notable elongation of particles near the surface (average particle aspect ratio between 0.45 and 0.55), attributed to plastic deformation under impact. An increase of the particles hardness especially at the coating surface by up to 50% was measured, attributed to work hardening and grain refinement confirmed by XRD and EBSD analysis. Synchrotron XRD measurements indicated low in-plane residual stresses in the aluminum phase ( -10< - 10 < σ < 0 MPa). Salt spray corrosion tests revealed that only the fully compacted coatings achieved the target exposure time. Under optimal process parameters, a continuous aluminum layer was formed, enabling effective sacrificial protection of the underlying steel substrate via aluminum particles consumption
Intégration de l'expertise militaire, des systèmes d'information géographique et de l'apprentissage automatique pour le déminage post-conflit : une approche basée sur la connaissance
No full textThe clearance of landmines and unexploded ordnance (UXO) in post-war environments remains a critical humanitarian and security challenge. Traditional demining efforts rely heavily on resource-intensive manual surveys and technological detection methods, yet they often suffer from inefficiencies due to incomplete spatial intelligence and environmental complexities. This research integrates Geographie Information Systems (GIS) and Machine Learning (ML) to enhance the prioritization and accuracy of landmine clearance operations. A novel approach is proposed that incorporates military expert knowledge into ML-driven risk assessment models, enabling more informed and adaptive decision-making. By leveraging spatial data, historical contamination records, terrain analysis, and remotely sensed imagery, the developed System optimizes hazardous area prediction, reducing uncertainty in the identification of mine- contaminated zones. The methodology is validated using real-world case studies from post-conflict regions, demonstrating its effectiveness in improving operational efficiency and safety. The results indicate that integrating domain expertise with data-driven intelligence enhances predictive performance, offering a scalable and adaptable framework for future demining efforts. This work contributes to the growing field of Al-assisted humanitarian operations, providing a robust foundation for more effective post-war landmine mitigation strategies.Le déminage des mines terrestres et des munitions non explosées (UXO) dans les environnements post-conflit demeure un défi humanitaire et sécuritaire majeur. Les méthodes traditionnelles de déminage reposent largement sur des relevés manuels coûteux en ressources et des technologies de détection, mais elles souffrent souvent d’inefficacités dues à un manque d’intelligence spatiale complète et à la complexité des environnements. Cette recherche intègre les Systèmes d’information Géographique (SIG) et l’Apprentissage Automatique (AA) afin d’améliorer la priorisation et la précision des opérations de déminage. Une approche novatrice est proposée, incorporant les connaissances d’experts militaires dans des modèles d’évaluation des risques basés sur l’AA, permettant ainsi une prise de décision plus éclairée et adaptative. En exploitant des données spatiales, des archives de contamination historique, des analyses topographiques et des images satellitaires, le système développé optimise la prédiction des zones à risque, réduisant ainsi l’incertitude liée à l’identification des zones potentiellement minées. La méthodologie est validée à travers des études de cas réelles issues de régions post-conflit, démontrant son efficacité dans l’amélioration de la sécurité et de l’efficacité opérationnelle. Les résultats indiquent que l'intégration de l’expertise du domaine avec l’intelligence basée sur les données améliore la performance prédictive, offrant ainsi un cadre évolutif et adaptable pour les futurs efforts de déminage. Cette recherche contribue au développement des opérations humanitaires assistées par l'IA et constitue une base solide pour des stratégies de mitigation des mines terrestres plus efficaces
Investigation of a Transonic Dense Gas Flow Over an Idealized Blade Vane Configuration
No full textInternational audienceAbstract An experimental and numerical study of the flow of the organic vapor Novec 649 flow and air through an idealized vane configuration was conducted. Due to the profiling of the passage, typical pressure, and Mach number distributions for transonic and supersonic flow around a turbine blade could be established. The comparable large trailing edge radius enabled the detailed investigation of corresponding flow phenomena. Schlieren pictures and pressure distributions were experimentally obtained and compared with numerical simulations. The trailing edge flow and the transonic and supersonic shock patterns were qualitatively similar to the ones observed in the perfect gas flow. However, significant quantitative deviations were observed. The base pressure was lower in dense gas flows characterized by a fundamental derivative below unity, Γ < 1, and the corresponding loss contribution was stronger
Reconstruction de modèles CAO paramétriques 3D basés sur l'apprentissage automatique à partir de dessins 2D
No full text3D reconstruction is a significant research area that has received much attention from academic and industry. Reconstructing parametric CAD models from 2D drawings is a complex and time-consuming task. This thesis concentrates on automatically and efficiently reconstructing 3D parametric CAD models from 2D drawings. Through a systematic literature review, this project is divided into three research problems: representation, learning, and optimization. We explore three different reconstruction strategies. First, a rule-based approach is proposed to convert 2D drawings into B-Rep models. Next, a deep learning-based approach converts dumb B-Rep models into parametric models. This reconstruction strategy implements the reconstruction pipeline with dumb B-Rep models as intermediate results. The third reconstruction strategy is an optimization of the first one to address the shortcomings of the first strategy. This approach directly converts 2D drawings into 3D parametric CAD models based on reinforcement learning without the need for intermediate results. The reconstructed results demonstrate good performance in both qualitative and quantitative comparisons. Moreover, this work is the first systematic study on reconstructing 3D CAD parametric models from 2D drawings.La reconstruction 3D est un domaine de recherche important qui a reçu beaucoup d'attention de la part des universités et de l'industrie. La reconstruction de modèles CAO paramétriques à partir de dessins 2D est une tâche complexe qui prend du temps. Cette thèse se concentre sur la reconstruction automatique de modèles CAO paramétriques 3D à partir de dessins techniques 2D. Après une revue systématique de la littérature, ce projet est divisé en trois problèmes de recherche : la représentation, l'apprentissage et l'optimisation. Nous explorons trois stratégies de reconstruction différentes. Tout d'abord, une approche basée sur des règles est proposée pour convertir les dessins techniques 2D en modèles B-Rep. Ensuite, une approche basée sur l'apprentissage profond convertit les modèles B-Rep en modèles paramétriques éditables. Cette stratégie de reconstruction met en œuvre le pipeline de reconstruction avec des modèles 3D B-Rep comme résultats intermédiaires. La troisième stratégie de reconstruction est une optimisation de la première afin de remédier aux lacunes de cette dernière. Cette approche convertit directement les dessins techniques 2D en modèles CAO paramétriques éditables basés sur l'apprentissage par renforcement sans avoir besoin de résultats intermédiaires. Les résultats reconstruits démontrent de bonnes performances dans les comparaisons qualitatives et quantitatives. En outre, ce travail est la première étude systématique sur la reconstruction de modèles CAO paramétriques en 3D à partir de dessins en 2D
Size effects in metallic polycrystals in the context of strain-integral crystal plasticity
No full textInternational audienceThe development of constitutive models usually relies on the framework of strain-gradient plasticity to consider the size and gradient effects that affect the thermomechanical behavior of crystalline materials. In this work, an alternative strategy, which fits into the category of strain-integral plasticity models, is explored. The underlying idea consists of evaluating the spatial average and the spatial covariance of the plastic deformation gradient tensor. These non-local variables are treated as additional internal state variables that provide some information regarding the spatial distribution of the plastic deformation gradient tensor.In the present paper, the method used for the evaluation of the average and the covariance of the plastic deformation gradient tensor is first detailed. Particular attention is paid to the treatment of near-boundary regions, for which different options are proposed. Then, a general strategy to include the average and the covariance of the plastic deformation gradient tensor in constitutive relations in a thermodynamically consistent manner is exposed. Finally, a crystal plasticity-based model developed within the framework of strain-integral plasticity is presented for the purpose of illustration. The numerical results obtained for different polycrystalline microstructures indicate that the hardening behavior is impacted by the mean grain size. However, such a size-dependent behavior largely depends on the method used for the treatment of near-boundary regions.</div
Enhancing Recruitment Selection Process In Human Resource With Artificial Intelligence Powered Resume Parser
No full textInternational audienceThe traditional recruitment process suffers from inefficiencies, subjective evaluations and biases that hinder hiring quality and increase costs. This research develops an AIpowered Resume Parser to automate resume screening process and address these challenges. The system uses Natural Language Processing (NLP) and machine learning techniques for automation, utilizing an XGBoost classifier for job categorization and a customized Named Entity Recognition (NER) model for information extraction. The models were trained on 2484 resumes across 24 job categories with comprehensive data pre-processing and addressed class imbalance using SMOTE. The performance of six algorithms was compared throughout the research, revealing that XGBoost achieved optimal accuracy of 77% in job categorization, outperforming other tested algorithms including Random Forest (68%), Support Vector Machine (SVM) (62%), and Logistic Regression (64%) etc. The customized NER model successfully extracted key entities including candidate name and skills with high accuracy in test cases, proving its effectiveness in resume parsing. This system revolutionizes the recruitment process by promoting fairness, efficiency and transparency while addressing the limitations of manual recruitment methods