HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
A Comprehensive Degradation Modelling: from Statistical to Artificial Intelligence Models
No full textInternational audienceIn the real world, a product or a system usually loses its function gradually with a degradation process rather than fails abruptly. To meet the demand of safety, productivity, and economy, it is essential to monitor the actual degradation process and predict imminent degradation trends. A degradation process can be affected by many different factors.Degradation modelling typically involves the use of mathematical models to describe the degradation processes that occur in materials or systems over time. These models can be based on empirical data, physical principles, or a combination of both, and can be used to make predictions about the future performance of the material or system.This work is attempted to review previous degradation models, and present some new deep learning based approaches for degradation modelling. First, it deals with statistical models, like general path and stochastic models. Then, because of some cumbersomeness of statistical models; like incompleteness modelling, it moves to make comforts by some AI models.The main advantage of AI models is capturing possible nonlinearity in the observed degradation data, but they often suffer from limitations of available dataset.To overcome limitations of statistical and machine learning models, some mixed models considering both simultaneously have been presented.This work is aimed at explaining briefly all models and then making a huge comparison between them for some irregular real degradation data. The mentioned data is related to the wear of some chains producing glass wool
Planification de la production pour la customisation de masse dans un système de fabrication reconfigurable
No full textIndustries nowadays are facing difficult socio-economic environment due to more demanding customers. To facilitate th increased diversity of customized products and Iow cost of mass production, reconfigurable manufacturing system (RMS is greatly recommended as the most suitable manufacturing system to mass customization (MC). The production planning of MC is complex. The reconfigurability of RMSs provides flexibility for MC without losing the economies of scal and efficiency of mass production. ln spite of that, the frequent machine and layout reconfigurations take time and mak MC production more complicated. Based on the above, the research in this thesis built multi-objective mathematical models integrating process planning and scheduling for MC as well as layout reconfiguration of RMS, in order to get an optimal production plan. This thesis first introduces a bi-objective model and the approximate solution approaches combined With the proposed constraint handling techniques. Then, it introduces a three-objective model extended from the initial bi-objective model by additionally considering environmental sustainability and presents a problem-specific modified Pareto efficiency for solution approaches. Finally, it introduces the developed Q-learning based reinforcement learning approach to solve the reduced bi-objective mathematical model.De nos jours, les industries sont confrontées à un environnement socio-économique difficile en raison de clients plus exigeants. Pour faciliter la diversité accrue des produits personnalisés et le faible coût de la production de masse, le système de fabrication reconfigurable (RMS) est fortement recommandé comme le système de fabrication le plus adapté à la personnalisation de masse (MC). La planification de la production de la personnalisation de masse est complexe. La reconfigurabilité des systèmes de fabrication reconfigurables offre une certaine souplesse pour MC sans perdre les économies d'échelle et l'efficacité de la production de masse. Malgré cela, les reconfigurations fréquentes des machines et de l'agencement prennent du temps et compliquent la production de MC. Sur la base de ce qui précède, la recherche dans cette thèse a construit des modèles mathématiques multi-objectifs qui intègrent la planification du processus et l'ordonnancement pour MC ainsi que la reconfiguration de l'agencement de RMS, afin d'obtenir un plan de production optimal. Cette thèse présente d'abord un modèle mathématique bi-objectif et les approches de solutions approximatives combinées aux techniques de traitement des contraintes proposées. Ensuite, elle introduit un modèle mathématique à trois objectifs étendu à partir du modèle bi-objectif initial en considérant en plus la durabilité environnementale, et présente une efficacité de Pareto modifiée spécifique au problème pour les approches de solution. Enfin, il présente l'approche d'apprentissage par renforcement basée sur l'apprentissage Q pour résoudre le modèle mathématique bi-objectif réduit
Revisiting organosolv strategies for sustainable extraction of valuable lignin: the CoffeeCat process
No full textInternational audienceAn innovative and sustainable strategy for the selective extraction of lignin from lignocellulosic biomass has been designed, namely the CoffeeCat process, in which only green solvents and reagents are required: (i) water, (ii) 2-methyltetrahydrofuran-3-one (coffee furanone) recognized as a food grade ingredient and readily biodegradable and (iii) glutamic acid. Two fractions have been isolated from Miscanthus x giganteus, the delignified fraction (DL-glu) and the enriched-lignin fraction (L-glu). Competitive extraction yields of 27% and 43% of enriched-lignin fractions were respectively obtained at 140 °C (Lglu-140) and 180 °C (L-glu-180), based on the lignin content of the original biomass. The structural properties of these lignins were characterized by spectroscopic (FTIR and NMR), microscopic (SEM) and separative (SEC-MALLS) methods. Compared to other processes described in the literature, our strategy involved the isolation of lignin fractions with high purity (up to 84%). Both fractions have been valorized: (i) the DL-glu fractions have been subjected to ionic liquid pretreatment and subsequent enzymatic hydrolysis leading to a total depolymerization of the constitutive cellulose (99%) and to an efficient conversion of hemicellulose into xylose (70%); (ii) the L-glu fractions have been used to produce lignin nanoparticles (LNPs) in a mixture of 2-methyltetrahydrofuran-3-one/water (1/110 v/v). The size distribution (272 ± 9 nm and 472 ± 6 nm), charge (−29.2 ± 0.8 mV and −20.8 ± 0.4 mV) and regular spherical shape of these LNPs have been determined using Zetasizer-DLS measurements and SEM images of L-glu-140 and L-glu-180, respectively. In addition, the possibility of easy incorporation of the L-glu fraction into polylactic acid without requiring previous lignin modification has been preliminarily explored. The CoffeeCat process was thus demonstrated as a relevant eco-solution for an integrated lignocellulosic biorefinery. Sustainability spotlight This work is based on the CoffeeCat process to design a sustainable strategy for a full-component renery of lignocellulose. The strategy uses green solvents (the food approved 2-methyltetrahydrofuran-3-one and water) and reagents (glutamic acid) to selectively extract at 140 °C a nearly carbohydrate-free lignin with high b-O-4 linkage content. Synergistic ionic liquid pretreatment applied to the remaining delignied biomass led to total conversion of polysaccharides into platform sugars. The physico-chemical properties of the extracted lignin could open the way of designing biocompatible nanoparticles and facilitating the formulation of PLA-lignin biocomposites. Miscanthus is used as an example for potential biomass feedstock. This work is in line with points 12 and 13 of the UN's Sustainable Development Goals (combat climate change and ensure sustainable production patterns)
Neutron spectrum unfolding using two architectures of convolutional neural networks
No full textInternational audienceWe deploy artificial neural networks to unfold neutron spectra from measured energy-integrated quantities. These neutron spectra represent an important parameter allowing to compute the absorbed dose and the kerma to serve radiation protection in addition to nuclear safety. The built architectures are inspired from convolutional neural networks. The first architecture is made up of residual transposed convolution's blocks while the second is a modified version of the U-net architecture. A large and balanced dataset is simulated following "realistic" physical constraints to train the architectures in an efficient way. Results show a high accuracy prediction of neutron spectra ranging from thermal up to fast spectrum. The dataset processing, the attention paid to performances' metrics and the hyperoptimization are behind the architectures' robustness
Digital commons governance.: How do Internet users govern themselves online?
No full textSeveral million volunteers have collaborated online around the Wikipedia project to write, in less than twenty years, the largest encyclopedia ever written by humanity. How did these volunteers, the vast majority of whom had never met each other, organize themselves to succeed in writing such a sum of knowledge made available to all, something that neither private publishers nor the most venerable public libraries had managed to do until then? This question is not limited to the case of Wikipedia. There are many examples of communities of Internet users self-organizing to collectively produce digital resources whose free access was guaranteed. This is the case of the OpenStreetMap geographic database, the Firefox browser, the VLC multimedia player or the Wikicar online maps. Researchers have described these resources as "digital commons", which combine these three dimensions: the contributory production of a resource, the right of access to which is equitably distributed, and the management of which by a community through self-governance. The purpose of this chapter is to characterize this last dimension of governance of the digital commons, surely the most difficult and controversial. Self-governance is not self-evident If the success of Wikipedia is no longer debated today, it was not self-evident at all at its launch. How could cooperation between strangers prevail over the conflicting, vandalizing or "free rider" behaviors that can occur around open access resources?Plusieurs millions de bénévoles ont collaboré en ligne autour du projet Wikipédia pour rédiger, en moins de vingt ans, la plus grande encyclopédie jamais écrite par l'humanité. Comment ces internautes bénévoles, dont la grande majorité ne se sont jamais rencontrés, se sont-ils organisés pour réussir à rédiger une telle somme de connaissance mise à disposition de tous, ce que ni les éditeurs privés ni les bibliothèques publiques les plus vénérables n'étaient parvenus à faire jusqu'alors ? Cette question ne se limite pas au cas de Wikipédia. Il existe de nombreux exemples de communauté d'internautes s'auto-organisant pour produire collectivement des ressources numériques dont le libre accès était garanti. C'est le cas de la base de données géographiques OpenStreetMap, du navigateur Firefox, du lecteur multimédia VLC ou encore des plans en ligne de la Wikicar. Les chercheurs ont qualifié de « communs numériques » ces ressources regroupant ces trois dimensions : production contributive d'une ressource, dont le droit d'accès est équitablement réparti et qu'une communauté gère en s'autogouvernant. Le propos de ce chapitre est de caractériser cette dernière dimension de gouvernance des communs numériques, sûrement la plus difficile et sujette à controverses. Une auto-gouvernance qui ne va pas de soi Si le succès de Wikipédia ne fait plus débat aujourd'hui, il n'allait pas du tout de soi à son lancement. Comment la coopération entre des inconnus pouvait l'emporter sur les comportements conflictuels, de vandalisme ou de « passager clandestin », qui peuvent avoir lieu autour des ressources en libre accès
An aftertaste of Cartesian salad? Pre-reflective self-consciousness, Peirce, and the study of cognition in the wild
No full textInternational audienceI situate the originality and the ambiguities of the target paper in the context of post-cognitivist cognitive science and in relation with some aspects of Charles Sanders Peirce’s anti-Cartesianism. I then focus on what the authors call « pre-reflective self-consciousness », highlighting some ambiguities of the characterizations they propose of this variety of consciousness. These ambiguities can become difficulties once one grants a crucial methodological role to this consciousness in the study of cognitive activities
Évaluation des dispositifs médicaux en vie réelle
No full textIn order to ensure optimal patient safety during care, medical devices must meet the performance and safety requirements set forth in Regulation (EU) 2017/745.Proof of compliance with these requirements is most often going to be provided by randomized clinical trials. Although these trials are the most widely used in the medical field due to the level of evidence they provide, their results can however be complemented by real-life data collected in routine care practice. Still not widely used in the field of medical devices, these data may, in the future, represent a major interest for manufacturers. It is in this context that this project was set up to inform medical device manufacturers as well as the various actors in this field (such as industries or researchers) about the importance of this real-life data and the way it is already used today. Emphasis was also placed on the regulatory aspect concerning the processing of this data, which constitutes so-called "sensitive" data and must therefore comply with the General Data Protection Regulation (GDPR).This dissertation proposes an inventory of the real-life data of existing medical devices, with a mapping to study these different data and their collection. This analysis is accompanied by issues to which these data can provide answers throughout the life cycle of the device.Afin d’assurer une sécurité optimale du patient lors de sa prise en charge, les dispositifs médicaux doivent répondre aux exigences de performance et de sécurité énoncées dans le règlement (UE) 2017/745.La preuve du respect de ces exigences va le plus souvent être apportée par des essais cliniques randomisés. Bien que ces essais soient les plus utilisés dans le domaine médical grâce au niveau de preuve qu’ils apportent, leurs résultats peuvent toutefois être complétés par des données de vie réelle récoltées en pratique de soins courants. Encore trop peu utilisées dans le domaine des dispositifs médicaux, ces données peuvent, à l’avenir, représenter un intérêt majeur pour les fabricants. C’est dans ce contexte que s’inscrit ce projet afin d’informer les fabricants de dispositifs médicaux ainsi que les différents acteurs de ce domaine (tels que les industries ou les chercheurs) quant à l’importance de ces données de vie réelle et de la manière dont elles sont déjà utilisées aujourd’hui. L’accent a également été mis sur l’aspect réglementaire concernant le traitement de ces données qui constituent des données dites « sensibles » et doivent de ce fait respecter le Règlement Général sur la Protection des Données (RGPD).Ce mémoire propose un état des lieux des données de vie réelle des dispositifs médicaux existants, avec une cartographie permettant d’étudier ces différentes données et leur collecte. Cette analyse est accompagnée de problématiques auxquelles ces données peuvent amener des éléments de réponse tout au long du cycle de vie du dispositif
Theoretical and experimental studies of chitin nanocrystals treated with ionic liquid or deep eutectic solvent to afford nanochitosan sheets
No full textInternational audienceChitin nanocrystals have gained growing interest due to their many excellent properties, however the higher crystallinity of these nano-rod-shaped particles is a major hindering factor in preparation of nanochitosans with low degree of acetylation (DA). Here we studied the effect of the ionic liquid 1-ethyl-3-methylimidazolium acetate (IL) and deep eutectic solvent (choline chloride:lactic acid) (DES) pretreatments of chitin nanocrystals (NCChits) before the deacetylation step using both experimental and theoretical approaches. The results showed that the ionic liquid pretreatment was able to partially disrupt the crystalline structure leading to a lower DA (18.2 %) after two cycles of deacetylation reaction. DES pretreatment, however, was unable to disturb the intra- and intermolecular hydrogen bonds, resulting in a DA of 73.6 % similar to that of unpretreated chitin nanocrystals (73.2 %). SEM images of chitin nanocrystals pretreated with ionic liquid showed that the crystals can rearrange into sheets. Molecular simulations reveal the detailed mechanism of chitin nanocrystal dissociation, in which the combination of a net molecular charge and hydrogen-bonding groups on a single scaffold (as is the case for [C2mim] and [OAc]) plays a paramount role. These results can open the way to afford controlled nanochitosan sheets from chitin nanocrystals
Valorisation de la biomasse : Extraction, caractérisation et hydrophobisation des hémicelluloses
No full textInternational audienc
L'échelle du quartier, un bon compromis pour la modélisation des déperditions thermiques ?
No full textInternational audienceLes déperditions thermiques de l’environnement bâti correspondentaux pertes de chaleur subies par un bâtiment. Ces pertes thermiquessont souvent causées par une isolation insuffisante. La limitation de sesdéperditions constitue l’un des principaux leviers de la sobriétéénergétique identifiés par les politiques publiques pour favoriser laterritorialisation de la transition énergétique. La modélisationthermique d'un environnement bâti s'effectue traditionnellement auniveau urbain, à partir de données d'entrée macro-économiques ou dela typologie des bâtiments, ou au niveau du bâtiment, à partir dedonnées physiques, empiriques ou statistiques. Entre ces deux échelles,l’échelle urbaine et l’échelle de l’environnement bâti, l’échelle duquartier pourrait produire des simulations relativement précises à uncoût raisonnable, en mobilisant une approche technique intégrée del’aménagement. Notre contribution présente les résultats d'une analysecomparative de trois méthodes (une au niveau urbain et deux au niveaudes quartiers) pour discuter de l’intérêt de privilégier les étudesthermiques à l’échelle du quartier, en vue de mieux diagnostiquer etainsi convaincre les usagers de réaliser des travaux de rénovationénergétique