HAL-MINES ParisTech
Not a member yet
27348 research outputs found
Sort by
Imaginaires sociaux des situations d'urgence en mer depuis 1800 : remarques introductives
No full textInternational audienc
Reporting de durabilité : rendre opérationnelle une construction théorique
No full textCSRD : quatre lettres qui ont bousculé les débats politiques européens cette année. Pour peut-être la première fois, un sujet comptable, réputé technique et aride, s’est invité dans les tribunes d’opinion et les quotidiens nationaux. La « Corporate Sustainability Reporting Directive » impose désormais aux grandes entreprises européennes de publier, en plus de leurs comptes financiers, des informations détaillées sur leurs impacts environnementaux, sociaux et de gouvernance.Sur le papier, difficile de s’opposer à une telle exigence. Face à l’urgence climatique et aux inégalités persistantes, il semble naturel d’exiger des entreprises qu’elles rendent des comptes : sur leurs émissions de CO₂, leur empreinte sur la biodiversité, leur politique de diversité ou encore leurs chaînes d’approvisionnement. Pourtant, la directive fait débat. Loin de trouver un consensus, elle cristallise les tensions : instrument de transparence ou usine à gaz réglementaire ? levier de transformation ou outil de communication verte ? Le débat est vif, et la Commission européenne a d’ailleurs annoncé une révision du texte à peine adopté.Dans ce mémoire, nous avons cherché à comprendre ce qui se joue réellement derrière le reporting de durabilité. Depuis octobre 2024, nous avons mené plus de cinquante entretiens avec les acteurs directement concernés. Et ils sont nombreux : entreprises de toutes tailles, investisseurs, auditeurs, ONG environnementales, syndicats, consommateurs, administrations, législateurs… Autant de parties prenantes qui préparent, utilisent, ou contestent les informations publiées.Notre objectif est double : éclairer les enjeux multiples et parfois contradictoires du débat sur la CSRD, et analyser les effets réels du reporting de durabilité. Quels mécanismes ce reporting active-t-il ? Quelles limites rencontre-t-il ? Peut-il être autre chose qu’une contrainte bureaucratique coûteuse ? Nous tentons ici de montrer comment le reporting de durabilité pourrait devenir un véritable outil de pilotage stratégique au service d’objectifs environnementaux et sociaux ambitieux, à condition d’éviter les écueils de la complexité et du conformisme passif
CAMS Solar Radiation day ahead forecasts: Site dependent evaluation of the IFS-COMPO forecast
No full textInternational audienceAs a part of the Copernicus Atmosphere Monitoring Service (CAMS), an hourly resolved 5 day forecast horizon irradiance forecast (IFS-COMPO) is provided regularly on the global scale. The particularity of this forecast is that It makes use of the CAMS forecast of the atmospheric composition, that is, aerosols, water vapor and ozone.It is an overarching question, whether the IFS-COMPO provides radiation forecasts with an added value compare to ECMWF’s operational IFS-HRES (high resolution) forecast and whether it can be used as an operational product on itself. The IFS-COMPO forecast is operated on a low spatial resolution of 40 km. This is coarse compared to the 9 km spatial resolution of the operational IFS-HRES run. Moreover, the IFS-COMPO runs uses a modeled aerosol forecasts from CAMS while the operational ECMWF IFS-HRES uses only an aerosol climatology. It is not obvious which of the two model runs provides the better radiation forecast quality.In this verification study an assessment on the years 2022and 2023 of the operational IFS-HRES forecast run in 9 km/1 hour spatio-temporal resolution and of the IFS-COMPO forecasts run in 40 km/1 hour spatio-temporal resolution is performed. The IFS-HRES run is used in its full spatial resolution but also spatially smoothed to a 40 km resolution which is directly comparable to the IFS-COMPO forecast resolution.As the reference data for this verification study, high-quality ground observations from various climate zones around the world were obtained for the same year. A total of 63 stations belonging to 7 different measurement networks were retrieved for the study. The ground stations retained were then classified into 5 site classes: continental, desert, mountain, subpixel and polar. The statistical error metrics mean bias error (MBE), mean absolute error (MAE) and root mean square error (RMSE) were processed separately for each class in all sky conditions and in cloud free (also called ‘clear-sky’ in solar energy applications) conditions.The main takeaways of the presentation are intended to be:1) to present the CAMS IFS-COMPO irradiance forecast to the community2) discuss recommendations for the solar energy community on how to use such a solar forecast based on an aerosol modeling approach
La justice organisationnelle passe-t-elle par la justesse de son instrumentation ? L’exemple de la répartition des ressources médicales au sein d’un département hospitalo-universitaire d’anesthésie-réanimation-médecine périopératoire pilotée par un calcul d’effectif
No full textInternational audienc
Resilience of global supply chains with logistics service uncertainty: Onshoring versus offshoring strategy
No full textInternational audienceSupply chain resilience aims to enhance supply chain’s capacity to resist, respond, and recover from the disruptions. In different industries, products exhibit distinct attributes, such as, product value, deterioration rate, and criticality, resulting in varied resilience performance under logistic disruptions. The goal of this study is to investigate the significance of logistics service network in enhancing resilience within different supply chain typologies (onshoring and offshoring) across diverse industries, in face to transit time uncertainty induced by disruptions. We develop a robust optimization model with budgeted uncertainty to address the involved single-commodity multimodal freight routing problem under transit time uncertainty. To manage real-world large-scale instances, we devised a tailored adaptive large neighborhood search algorithm and validated its performance by computational experiments. Furthermore, we conducted realistic numerical experiments to validate the proposed approach, and sensitivity analyses to assess supply chain resilience from two perspectives: supply chain typologies and industry-specific attributes including product value, deterioration, and criticality. Moreover, the dynamics of modal shift under various disruptions were also analyzed. The study aims to provide practical tools and managerial insights to help industrial practitioners improve the resilience of their logistics service networks and emphasizes the importance of industry-specific considerations of supply chain typology inenhancing resilience
From Knowledge Management to Data Management in Innovation: Organizing to Leverage Data Through Anomalies
No full textInternational audienceABSTRACT Owing to their wide accessibility, data are currently at the root of many opportunities and challenges. Among the opportunities is the value that data can generate for innovation and new product development. The managerial implementation of this so‐called value generation constitutes an associated challenge. This study addresses this issue with the goal of proposing a rationalisation of data management that allows organisations to leverage data for innovation. We draw upon knowledge management (KM) literature because data and knowledge have been closely linked for decades, as seen in the knowledge pyramid linking data to wisdom. Akin to KM practices, we begin by uncovering the four necessary dimensions that must be structured to leverage data (generation, relations, usage, and technologies). By recalling the singularities of data compared with knowledge, we hypothesise that anomalies play a potential role in operationalizing this framework for value generation. Evidence from several case studies support the proposed framework and lead us to introduce the notion of data heritage as a necessary condition for initiating a value generation process from data. Second, we highlight the importance of coupling between this data heritage and the knowledge heritage of organisations. Third, we emphasise that this coupling can be effectively realised through anomalies. Consequently, in response to recent calls in innovation management literature, this study provides insights into value generation from data by indicating several necessary conditions that need to be fulfilled. Moreover, it also benefits the KM literature by stressing the logic of a new KM pyramid
Contributions à la détection non supervisée d'anomalies visuelles
No full textThis thesis addresses visual anomaly detection, focusing on unsupervised defect identification in industrial inspection applications. Motivated by recent advances in the field, this work presents contributions to (1) the evaluation standards of anomaly localization, (2) usability of models via post-processing techniques, and (3) model-specific improvements. First, we introduce AUPIMO, a novel evaluation metric that addresses limitations of existing benchmarks. It imposes a hard minimization of false positives on normal samples, encouraging a more challenging and trustworthy evaluation. Experiments across 27 datasets and eight state-of-the-art models demonstrate that AUPIMO provides a more reliable and detailed performance assessment than previous benchmarks. Second, we propose an unsupervised, image-specific threshold selection method for anomaly localization. This method avoids biases introduced by statistics-based thresholds, offering an alternative to color-coded heatmaps. Third, we analyze Gaussian distribution-based models through a novel dimensionality reduction method. This analysis leads to findings that challenge prevailing notions in the field, such as the correlation between variance and model performance. This novel subspace-based dimensionality reduction method, combined with synthetic anomalies, is shown to consistently improve performance. Additionally, we introduce a visualization tool enabling anomaly localization for image-wise Gaussian models. The thesis also presents an incremental improvement to a pixel-wise one-class classification model, enabling more effective use of pixel-wise annotations and faster training. The proposed contributions aim to bridge the gap between research and real-world applications, offering model-agnostic solutions to improve benchmarking and model usability, and model-specific improvements to Gaussian-based models.Cette thèse traite de la détection d'anomalies visuelles, en particulier appliquée à l'identification non supervisée de défauts en applications d'inspection industrielle. Motivé par les récentes avancées dans le domaine, ce travail présente des contributions (1) à d'évaluation de la localisation des anomalies, (2) à la facilité d'utilisation des modèles via des techniques de post-traitement, et (3) à des améliorations spécifiques à des modèles basés sur les distributions gaussiennes. D'abord, nous présentons AUPIMO, une nouvelle mesure d'évaluation qui s'attaque aux limites des emph{benchmarks} actuels. Elle impose une minimisation stricte des faux positifs sur les échantillons normaux, encourageant ainsi une évaluation plus difficile. Des expériences menées sur 27 ensembles de données et huit modèles de l'état de l'art démontrent qu'AUPIMO fournit une évaluation plus fiable et plus détaillée des performances. Deuxièmement, nous proposons une méthode de sélection non supervisée de seuil et spécifique à l'image pour la localisation des anomalies. Cette méthode évite des biais introduits par les seuils basés sur des statistiques et offre une alternative aux emph{heatmaps} codées en couleur. Troisièmement, nous analysons les modèles basés sur distributions gaussiennes à l'aide d'une nouvelle méthode de réduction de la dimensionnalité. Cette analyse aboutit à des résultats qui remettent en question des notions dominantes dans le domaine, telles que la corrélation entre la variance et la performance en détection d'anomalies. Cette nouvelle méthode de réduction de la dimensionnalité basée sur les sous-espaces, combinée à des anomalies synthétiques, permet d'améliorer les performances de manière consistante. Nous présentons également un outil de visualisation permettant de localiser les anomalies pour les modèles gaussiens conçu pour des images entières. La thèse présente également une amélioration incrémentale d'un modèle de classification à une classe, permettant une utilisation plus efficace des annotations au niveau des pixels et un entrainement plus rapide. Les contributions proposées visent à combler l'écart entre la recherche et les applications du monde réel, en offrant des solutions agnostiques aux modèles pour améliorer l'analyse comparative et la facilité d'utilisation des modèles, ainsi que des améliorations spécifiques aux modèles basés sur les gaussiennes
Le patrimoine industriel : Un levier stratégique pour la réindustrialisation et la souveraineté économique
No full textInternational audienc
The (Re)Design of Ecosystems to Face Grand Challenges—Toward the Management of Creative Evolution
No full textInternational audienceABSTRACT Today's grand challenges have to be discussed comprehensively, addressing the current design and the redesign of social and economic ecosystems. Tackling grand challenges requires a step into the unknown, a situation that is well‐characterised today as strongly different in nature from uncertainty and requires specific types of management. This opens the door to radical changes in ecosystems' configurations (new values, new interdependencies and new independencies). Hence, our call for papers for this special issue asked about (re)designing ecosystems to face grand challenges. The papers of this special issue analyse a great variety of empirical situations (energy, organised crime, wind‐energy–air quality). They elaborate extensively on how one can manage the complex intertwined processes that lead to exploring both new products/services associated with grand challenges and new collaborations in the ecosystem, enabling the commitment of new players in new partnerships. This editorial highlights three pivotal constructs deriving from the contributions: (A) ‘generativity enhancer’ as a critical type of action/actor in the redesign of the ecosystem; (B) creative preservation as a critical factor shaping performance; and (C) necessary methods for creating/sparing effective multiple local design spaces. Building on the constructs, it concludes by discussing the (re)design of ecosystems as creative evolution and how to manage it
