Portail HAL IMT Mines Albi
Not a member yet
5440 research outputs found
Sort by
Influence of the scanning strategy on the microstructure and the tribological behavior of a Ni-based superalloy processed by L-PBF additive manufacturing
No full textInternational audienceThe ABD-900AM is a newly developed nickel-based alloy specifically designed for additive manufacturing, and it can be printed using a wide range of process parameters. This alloy, combined with the L-PBF process, allows for the production of dense parts with reference microstructures that are multi-scale, and facilitates the study of the links between these microstructures and their tribological behaviors. To study the relationships between microstructures and tribological behaviors, the parts are built vertically without interlayer rotation.The plane studied are obtained with three scanning directions : 0◦, 45◦ and 90◦. These angles allow to generate different microstructures on the planes on which are conducted the microstructural characterizations and tribological tests, for which the sliding direction is always parallel to the building direction. This choice of parameter enables access to different crystallographic textures (200) and (111), and cellular structures with varying morphologies and homogeneity within the melt pools. Tribological tests were performed on these microstructures in a ball-on-flat configuration with alternating motion, without lubrication, and at room temperature. For the three laser angles (0◦, 45◦, and 90◦), the wear volumes are 0.60 ± 0.12 mm3, 0.81 ± 0.15 mm3, and 1.00 ± 0.09 mm3, respectively. The wear resistance is primarily related to the amount of oxide layers formed on the wear track of the samples
Selective laser melting of partially amorphous regolith analog for ISRU lunar applications
No full textInternational audienceAs the idea of crewed outposts on the Moon gains momentum, In-Situ Resource Utilization (ISRU) technologies tend to become imperative to fulfill astronauts' needs. This article explores a way to use the lunar regolith as a source material for the additive manufacturing of complex objects, based on the selective laser melting (SLM) technique. A lunar regolith analog, Basalt of Pic d’Ysson (BPY), is used as a starting point for this study, to investigate the now demonstrated impact of amorphous analog content in the powder bed, substrate type, and post-SLM annealing treatments on the mechanical properties of 3D-printed objects. Improvements to the manufacturing and sample extraction stages are proposed to systematically reproduce the high compressive strength values obtained, thus contributing to the robustness and reliability of the process
A semi-analytical model for low-density impact-based surface treatments: Application to the abrasive waterjet texturing of thermoplastic polymers
No full textInternational audienceSurface roughness is critical for bonding applications, as it directly influences the mechanisms occurring at the adhesive interface. Abrasive Waterjet texturing has emerged as a promising technique for functionalizing surfaces, but predicting the surface characteristics from stochastic impact-based processes remains a challenge. This study aimed to develop a numerical model capable of forecasting key morphological parameters for AWJ-textured surfaces with pilotable treatment coverage. The proposed model was optimized through theoretical analysis and confronted to topographical data from polymer samples treated with low-density AWJ using standard parameters. Profilometry measurements were supported by a custom post-treatment algorithm to remove artefacts and assess the characteristics of individual particle impacts (number, repartition, dimensions). The predicted roughness showed a 94 % concordance to the measured values
Outil d'aide à la décision par apprentissage automatique pour la planification intégrée de centres d'appels soumis à des incertitudes : cas d'étude d'un centre relais pour sourds et malentendants
No full textWith the call center sector on the rise, the importance of focusing research efforts towards optimizing and improving this sector is clear. The performance of a call center is directly linked to the quality of agent schedules being executed on a daily basis, with 60%-70% of call center operating expenses being attributed to human resource costs as seen throughout the literature. In order to both ensure the accessibility of callers and attend to the call center's managerial and agent requirements, these agent schedules must be optimized. However, the call center environment is one of stochastic nature which makes it all the more difficult to manage. The planning supervisor, a representative charged with the staffing and scheduling task within the call center, must account for this stochasticity when assigning agents throughout the planning period. This stochasticity is represented by unknown volumes of incoming calls, variable call handling times and a significant human factor. Various domains of research including forecast generation, queueing theory and simulation demonstrate interest in the different aspects of the call center environment. Despite the extensive historical literature regarding agent staffing and scheduling within call centers, the vast majority of carried out studies rely on performance evaluation models that date back to the early and mid-20th century. Moreover, due to the non-conventional nature of modern call centers, the assumptions on which these classical approaches are based—and, consequently, the performance of these models in generating agent schedules—are called into question. In this Ph.D. thesis, we direct our attention to solving the agent staffing and scheduling problem using advanced machine learning-based techniques with our study case being a relay call center for the deaf and hearing-impaired community. This prompts the formulation of the following research questions: (1) How can we develop an optimized agent schedule in light of the existing uncertainties using a machine learning-based approach? (2) How can we validate the performance of our proposed approach? The contribution of this manuscript is twofold: (1) A scientific contribution in the form of a machine learning-based approach for the integrated staffing and scheduling of agents. An artificial neural network model is used for performance evaluation within a deep reinforcement learning-based framework for agent schedule generation. (2) A technical contribution in the form of a complete pipeline starting from the processing of available call center historical data and ending with the generation of agent schedules. Using this pipeline, the performance of our proposed approach is then compared against that of more classical approaches. Finally, leveraging the implementation of our proposed scientific approach within the mentioned pipeline, it is made possible to construct a decision support system dedicated to call center planning supervisors. Functioning within the context of uncertainty, such a system would propose proficient agent schedules that satisfy the set of performance criteria set forth by the call center management.Les centres d'appels étant en plein essor, il est important de concentrer les efforts de recherche sur leur optimisation et leur amélioration. La performance d'un centre d'appel est directement liée à la qualité de la planification des agents. En effet, 60% à 70% des dépenses d'exploitation des centres d'appels sont attribuées aux coûts des ressources humaines. Afin de garantir l'accessibilité des appelants et de répondre aux besoins des gestionnaires et des agents du centre d'appel, la planification des agents doit être optimisée. Cependant, l'environnement d'un centre d'appel est de nature stochastique, ce qui le rend d'autant plus difficile à gérer. Cela se manifeste par des volumes d'appels entrants inconnus, des temps de traitement d'appels variables, et un facteur humain important. Le superviseur, chargé des tâches de planification au sein du centre d'appel, doit prendre en compte cette stochasticité lors de l'affectation des agents. Divers domaines de recherche comme la théorie des files d'attente, la prédiction et la simulation, ont montré leur intérêt pour traiter les différents problèmes relatifs aux centres d'appels. Malgré l'abondante littérature scientifique concernant la planification des agents dans les centres d'appels, la grande majorité des études réalisées s'appuient toujours sur des modèles d'évaluation de performance qui datent du début et du milieu du XXe siècle. De plus, en raison de la nature non conventionnelle des centres d'appels modernes, les hypothèses sur lesquelles ces approches classiques sont basées - et, par conséquent, la performance de ces modèles dans la génération de plannings - sont remises en question. Dans cette thèse de doctorat, nous nous intéressons à la résolution du problème de planification des agents en utilisant des techniques avancées basées sur l'apprentissage automatique. Notre cas d'étude se concentre sur un centre d'appel relais dédié à la communauté des sourds et malentendants. Cela nous amène à formuler les questions de recherche suivantes : (1) Comment développer un planning optimisé en tenant compte des incertitudes existantes et en utilisant une approche basée sur l'apprentissage automatique ? (2) Comment valider les performances de l'approche proposée ? La contribution de ce manuscrit est double : (1) Scientifique, sous la forme d'une approche intégrée basée sur l'apprentissage automatique. Les plannings sont générés à l'aide d'une technique d'apprentissage par renforcement, et leur performance est évaluée grâce à un réseau de neurones. (2) Technique, sous la forme d'un processus complet commençant par le traitement des données historiques disponibles du centre d'appel et se terminant par la génération de plannings. Ce processus permet notamment d'évaluer les performances de notre approche comparativement aux approches classiques. Finalement, ces deux contributions permettent de proposer un système d'aide à la décision destiné aux superviseurs des centres d'appels en leur proposant des plannings satisfaisants les critères de performance souhaités dans un contexte incertain
Rolling horizon data driven robust optimization for supply chain planning
No full textInternational audienceThis paper discusses the challenge of production planning in a dyadic supply chain, where uncertainties disrupt plans that are updated through a rolling horizon DRP process. These uncertainties, such as demand fluctuations, machine failures, and delivery delays, cause instability and worsen the bullwhip effect, which reduces the reliability and effectiveness of production plans.To address these challenges, we propose a data-driven robust optimization framework that uses historical data analysis and clustering techniques to create well-defined uncertainty sets. Numerical experiments, using simulated historical data, show that this approach improves supply chain planning by balancing precision and robustness in managing uncertainties
Identification of the Potential Contribution of the Physical Internet to the Issues and Challenges of the Coffee Supply Chain in Ethiopia
No full textInternational audienceOur article explores the challenges and problems associated with managing the coffee supply chain in developing countries and examines how the physical Internet can solve them. To do this, we look at the case of the Ethiopian coffee supply chain. By leveraging hyperconnected logistics, multimodal transport and digital supply chain solutions, we highlight opportunities to improve economic, environmental and societal sustainability. Our literature review identifies key barriers such as supply chain fragmentation, lack of real-time visibility and inefficient transport and storage. We propose that the principles of the physical Internet should be harnessed to improve traceability, reduce waste and promote fair trade, particularly in the coffee supply chain, where they have not been used before. This research contributes to the wider debate on sustainable, resilient and inclusive global supply chains
Utilizing large pre-trained monocular depth models for machining setup inspection and measurement
No full textInternational audienceMonocular depth estimation has many potential uses in industrial inspection, including pose estimation, conformity checks and volumetric measurements. However, producing accurate depth maps is a challenging task, especially in production of mechanical parts, where many reflective metallic surfaces are encountered. Additionally, industrial use cases often involve working with a low volume of available data, making it difficult to train custom deep learning models for depth estimation. As a solution, this paper demonstrates the viability of pre-trained monocular depth models for inspection of 5-axis machining setups. We showcase that outputs of such deep learning models can be adjusted to specific settings, based on known values of 5-axis machine coordinates. Moreover, this can be done in a way that avoids fine-tuning the network parameters using a large amount of custom data
Vers la simulation numérique des pores dans des pièces composites stratifiées
No full textInternational audienceCes travaux visent à anticiper les futurs défis liés à la porosité dans les composites préimprégnés thermodurcissables. Bien que la maîtrise de la porosité soit actuellement satisfaisante, l'évolution des enjeux et des besoins dans le domaine de l'aéronautique pourrait entraîner l'apparition de nouveaux phénomènes de porosité, compromettant les performances mécaniques et la durabilité des structures. La formation de pores résulte principalement de l'air piégé, des gaz dissous dans la résine, des volatiles résiduels, et d'une imprégnation inadéquate des fibres. Une première implémentation du modèle de croissance de bulle développé dans une étude précédente [1] dans des sous-programmes numériques User MATerial (UMAT) pour le solveur éléments finis Abaqus a été corrigé et amélioré. La simulation de l'évolution du rayon de bulle au cours de la polymérisation montre de bons résultats par rapport aux mesures expérimentales. Cependant, ces modèles ne prennent pas encore en compte les interactions entre bulles et fibres, ni même entre deux bulles (coalescence). Pour pallier cela, des approches type fluides à l'échelle microscopique basées sur le logiciel open source OpenFOAM sont envisagées, ainsi que l'utilisation d'approches non-locales dans Abaqus. Les prochaines étapes de ces travaux de recherche consisteront à valider expérimentalement le modèle
In Situ Monitoring of Retained Austenite Decomposition During Tempering of High-Strength Tool Steels
No full textInternational audienceThis study investigates the decomposition of retained austenite (RA) in tool steels for plastic molding in correlation with the alloy chemical composition and the tempering parameters. Two grades differing in their silicon content with initial mixed bainitic/martensitic microstructures were investigated using in situ synchrotron high-energy X-ray diffraction (HEXRD) during tempering in the 550 °C to 600 °C temperature range for one-hour holding time. Results indicated carbide formation during heating or isothermal holding; however, retained austenite remained untransformed up to the end of the tempering holding time in all investigated conditions for both grades. In situ HEXRD provides direct evidence of the transformation of retained austenite into fresh martensite on cooling from the tempering stage. This behavior is correlated to the evolution of carbon enrichment of retained austenite and the effect of silicon is discussed
Machine learning-based agent staffing under uncertainty: The case of a relay call center
No full textInternational audienceClassical queueing models fail to properly staff non-conventional call centers with complex internal structures. This is either due to the difficulty of finding suitable models whose underlying assumptions hold, or due to certain elements of the call center not being modeled such as caller patience times. Relay call centers, service providers that connect two different interested parties with one another through telecommunication channels, present a prime example of non-conventional call centers. Working on the study case of a relay call center for the deaf community, Erlang C, one of the most commonly used call center staffing formulae, fails to generate agent staffing that meets our target performance criteria for quality of service. We propose a machine learning-based approach leveraging an available log of historical data. Upon comparing the proposed approach’s capability of performance evaluation and agent staffing to that of the Erlang C model and a baseline data-driven model, results indicate our approach’s staffing superiority. Considering uncertainty within the system variable predictions carried out prior to the staffing phase, our approach generates agent staffing which enables us to meet our global quality of service objective