HAL Arts et Métiers
Not a member yet
14127 research outputs found
Sort by
Am I (Not) a Ghost? Leveraging Affordances to Study the Impact of Avatar/Interaction Coherence on Embodiment and Plausibility in Virtual Reality
No full textInternational audienceThe way users interact with Virtual Reality (VR) environments plays a crucial role in shaping their experience when embodying an avatar. How avatars are perceived by users significantly influences their behavior based on stereotypes, a phenomenon known as the Proteus effect. The psychological concept of affordances may also appear relevant when it comes to interact through avatars and is yet underexplored. Indeed, understanding how virtual representations suggest possibilities for action has attracted considerable attention in the human-computer interaction community, but only few studies clearly address the use of affordances. Of particular interest is the fact aesthetic features of avatars may signify false affordances, conflicting with users' expectations and impacting perceived plausibility of the depicted situations. Recent models of congruence and plausibility suggest altering the latter may result in unexpected consequences on other qualia like presence and embodiment. The proposed research initially aimed at exploring the operationalization of affordances as a tool to investigate the impact of congruence and plausibility manipulations on the sense of embodiment. In spite of a long and careful endeavor materialized by a preliminary assessment and two user studies, it appears our participants were primed by other internal processes that took precedence over the perception of the affordances we selected. However, we unexpectedly manipulated the internal congruence following repeated exposures (mixed design), causing a rupture in plausibility and significantly lowering scores of embodiment and task performance. The present research then constitutes a direct proof of a relationship between a break in plausibility and a break in embodimen
Experimental study of capillary impregnation and wettability effects in porous cotton fiber structures
No full textInternational audienceThe study of capillary flows in cellulose fibers is important for various applications, including biomass pyrolysis and drying processes. This work investigates the behavior of cotton fibers during capillary impregnation using a dynamic approach. The analysis utilizes the Washburn equation and tensiometric methods to investigate geometric factors, apparent advancing contact angles, surface free energy of cotton fiber, and capillary pressure.The research is carried out in two phases. The first phase focuses on the theoretical application of the Washburn equation in porous cotton fibers, specifically examining capillary wicking behavior within a cylindrical holder. The second phase involves experimental analysis, using three different liquids: n-heptane, water, and glycerol. The surface tension of the liquids was measured, and the capillary impregnation process was characterizedthrough the determination of geometric factors, apparent advancing contact angles, and surface free energy. The geometrical factors of cotton fibers within the sample holder were found to be 10.39 ± 1.28 mm5. The apparent advancing contact angles for water and glycerol were 74.93◦ ± 2.20◦ and 69.55◦ ± 1.83◦, respectively
An innovative protocol for on-field biomechanical analysis of the hammer throw
No full textInternational audienc
Open Review of "Increasing dynamic range of Nonlinear Energy Sinks by using geometric nonlinear damping"
No full textThis is the Open Review of article https://doi.org/10.46298/jtcam.1328
Non-uniform truss modelling and energy consumption in adaptive space lattice manufacturing for steel structures
No full textInternational audienceThe present paper studies the energetic footprint of Adaptive Space Lattice Manufacturing (ASLM), a new metallic-based AM process joining solid rods through laser spot welding. Associated energy costs are evaluated and compared with other metallic-based AM processes -machining, casting, powder-based AM and wire-based AM. Starting with measures of the energy consumption on site, a comprehensive study is performed across different units of measure (kWh/kg, kWh/rod, kWh/m3, kWh/I-beam). An energy breakdown is provided, including a material-process balance assessment through the integration of material embodied energy. Finally, the impact of design opportunities offered by ASLM is studied, evaluating various unit-cells lattice geometries across a sensitivity study. ASLM is found to achieve an 88,6-99,7% energy reduction compared to other AM technologies, and lattice configurations assessed feature 1:10 energy consumption differences
颤动时空中基本粒子的几何起源与有限分类
No full textDiese Arbeit zeigt, dass alle bekannten fundamentalen Teilchen als geometrische Eigenmoden des zitternden Raum-Zeit-Gefüges entstehen: lokalisierte, kausal stabile Schwingungen der Raumzeitmetrik selbst. Im Rahmen der Theorie der Zitternden Raumzeit-Relativität (TSRT) werden Teilchenidentität, Masse, Spin und Ladung nicht als unabhängige Annahmen eingeführt, sondern deterministisch als intrinsische Eigenschaften symmetriegebundener Abweichungen der zugrunde liegenden Geometrie abgeleitet. Aufbauend auf früheren grundlegenden Arbeiten identifiziert die TSRT genau vier kausal zulässige Symmetrieklassen — U(1), SU(2), SU(3) und gravitative Krümmung — die jeweils eindeutig einer der bekannten fundamentalen Wechselwirkungen entsprechen. Diese Klassifikation ist zugleich äußerst restriktiv und, innerhalb der analysierten Grenzen, empirisch vollständig: Sie reproduziert die Eichgruppen des Standardmodells und sagt — soweit keine weiteren kausal zulässigen zitternden Symmetrieklassen identifiziert wurden — das Fehlen zusätzlicher fundamentaler Kräfte oder Teilchengenerationen voraus. Im Gegensatz zur Quantenfeldtheorie, die beliebige Felderweiterungen erlaubt, schließt die TSRT Superpartner, zusätzliche Familien oder exotische Eichsektoren aus, sofern sie nicht als kausal kohärente metrische Abweichungen realisierbar sind — was sich in der gegenwärtigen Formulierung als unvereinbar mit der variationsprinzipiellen Struktur der Theorie erweist. Ladungsquantisierung, Spin-Statistik-Korrespondenz, Farbladungs-Konfinierung und die Diskretheit der Teilchenfamilien ergeben sich allesamt aus der kausalen Kohärenz und dem topologischen Abschluss zitternder Geodäten, ohne Rückgriff auf Feldquantisierung oder eine probabilistische Interpretation. Über die Klassifikation hinaus liefert die TSRT eine geometrische Grundlage für Prinzipien, die traditionell als axiomatisch gelten. Die Herleitung von Ruheenergie und träger Masse folgt direkt aus der Akkumulation der Wirkung entlang der Eigenzeit und ergibt die Einstein-Beziehung zwischen Energie und Masse als deterministisches Resultat des geodätischen Widerstands gegen Krümmung. Wechselwirkungen werden als krümmungsvermittelte Kopplungen zwischen zitternden Moden formuliert, die den Austausch virtueller Teilchen durch deterministische geometrische Korrelationen ersetzen. Das Higgs-Boson wird nicht als quantisierte Anregung eines skalaren Feldes, sondern als metastabile, krümmungsgebundene Resonanz interpretiert, die durch geometrische Fragmentation zerfällt. Schließlich liefert die TSRT empirische Vorhersagen, die sie vom Standardmodell unterscheiden, darunter geometrische Einschränkungen der Fermionengenerationen, minimale Wirkungsskalen, die mit der Planckschen Konstante verknüpft sind, sowie überprüfbare Signaturen krümmungsinduzierter Zitterbewegung. Indem sie das vollständige Spektrum der bekannten Teilchen und Wechselwirkungen aus ersten Prinzipien ableitet — und zugleich präzise Falsifikationsbedingungen angibt — fördert diese Arbeit das Programm der geometrischen Vereinheitlichung und zeigt, dass Materie aus einem einzigen grundlegenden Prinzip hervorgeht: der kausalen Kohärenz im zitternden Raum-Zeit-Gefüge.This paper demonstrates that all known fundamental particles arise as geometric eigenmodes of trembling spacetime: localized, causally stable oscillations of the spacetime metric itself. Within the framework of Trembling Spacetime Relativity Theory (TSRT), particle identity, mass, spin, and charge are not introduced as independent assumptions but emerge deterministically as intrinsic properties of symmetry-bound deviations in the underlying geometry. Building on earlier foundational work, TSRT identifies exactly four causally admissible symmetry classes—U(1), SU(2), SU(3), and gravitational curvature—each corresponding uniquely to one of the known fundamental interactions. This classification is both highly restrictive and, within the constraints analyzed, empirically complete: it recovers the Standard Model gauge groups and predicts, to the extent that no additional causally admissible trembling symmetry classes have been identified, the absence of further fundamental forces or generations. In contrast to quantum field theory, which permits arbitrary field extensions, TSRT excludes superpartners, extra families, or exotic gauge sectors unless they can be realized as causally coherent metric deviations—an outcome shown to be incompatible with the theory’s variational structure under the current formulation. Charge quantization, spin-statistics correspondence, confinement of color charge, and the discreteness of particle families all emerge from the causal coherence and topological closure of trembling geodesics, without recourse to field quantization or probabilistic interpretation. Beyond classification, TSRT provides a geometric foundation for principles traditionally regarded as axiomatic. The derivation of rest energy and inertial mass follows directly from proper-time action accumulation, yielding the Einstein relation between energy and mass as a deterministic outcome of geodesic resistance to curvature. Interactions are formulated as curvature-mediated coupling among trembling modes, replacing virtual particle exchanges with deterministic geometric correlations. The Higgs boson is reinterpreted not as a quantized scalar field excitation but as a metastable curvature-bound resonance decaying through geometric fragmentation. Finally, TSRT yields empirical predictions distinguishing it from the Standard Model, including geometric constraints on fermion generations, minimal action scales linked to Planck’s constant, and testable signatures of curvature-induced trembling. By deriving the complete set of known particles and interactions from first principles—while identifying precise conditions for falsification—this work advances the geometric unification program and shows that matter originates from a single underlying principle: causal coherence in trembling spacetime.Este artículo demuestra que todas las partículas fundamentales conocidas surgen como modos propios geométricos del espaciotiempo tembloroso: oscilaciones localizadas y causalmente estables de la propia métrica del espaciotiempo. En el marco de la Teoría de la Relatividad del Espaciotiempo Tembloroso (TSRT), la identidad de las partículas, su masa, espín y carga no se introducen como hipótesis independientes, sino que emergen de manera determinista como propiedades intrínsecas de desviaciones ligadas a simetrías en la geometría subyacente. Basándose en trabajos fundamentales previos, la TSRT identifica exactamente cuatro clases de simetría causalmente admisibles — U(1), SU(2), SU(3) y la curvatura gravitacional — cada una de las cuales corresponde de manera única a una de las interacciones fundamentales conocidas. Esta clasificación es al mismo tiempo extremadamente restrictiva y, dentro de los límites analizados, empíricamente completa: reproduce los grupos gauge del Modelo Estándar y predice, en la medida en que no se han identificado otras clases de simetría temblorosa causalmente admisibles, la ausencia de fuerzas fundamentales adicionales o generaciones extra. En contraste con la teoría cuántica de campos, que permite extensiones arbitrarias de los campos, la TSRT excluye supercompañeros, familias adicionales o sectores gauge exóticos, a menos que puedan realizarse como desviaciones métricas causalmente coherentes — una posibilidad que resulta incompatible con la estructura variacional de la teoría en su formulación actual. La cuantización de la carga, la correspondencia entre espín y estadística, el confinamiento de la carga de color y la discreción de las familias de partículas surgen todas de la coherencia causal y del cierre topológico de las geodésicas temblorosas, sin recurrir a la cuantización de campos ni a una interpretación probabilística. Más allá de la clasificación, la TSRT proporciona una base geométrica para principios tradicionalmente considerados axiomáticos. La derivación de la energía en reposo y la masa inercial surge directamente de la acumulación de la acción en el tiempo propio, produciendo la relación de Einstein entre energía y masa como un resultado determinista de la resistencia geodésica a la curvatura. Las interacciones se formulan como acoplamientos mediados por la curvatura entre modos temblorosos, reemplazando el intercambio de partículas virtuales por correlaciones geométricas deterministas. El bosón de Higgs se reinterpreta no como una excitación cuantizada de un campo escalar, sino como una resonancia metastable ligada a la curvatura que decae mediante fragmentación geométrica. Finalmente, la TSRT proporciona predicciones empíricas que la diferencian del Modelo Estándar, incluyendo restricciones geométricas sobre las generaciones de fermiones, escalas mínimas de acción vinculadas a la constante de Planck y firmas comprobables del temblor inducido por la curvatura. Al derivar el conjunto completo de partículas e interacciones conocidas a partir de primeros principios — e identificar condiciones precisas de falsación — este trabajo impulsa el programa de unificación geométrica y demuestra que la materia procede de un único principio fundamental: la coherencia causal en el espaciotiempo tembloroso.Cet article démontre que toutes les particules fondamentales connues apparaissent comme des modes propres géométriques de l’espace-temps tremblant : des oscillations localisées et causalement stables de la métrique de l’espace-temps elle-même. Dans le cadre de la théorie de la relativité de l’espace-temps tremblant (TSRT), l’identité des particules, leur masse, leur spin et leur charge ne sont pas introduits comme des hypothèses indépendantes, mais émergent de manière déterministe en tant que propriétés intrinsèques des déviations liées aux symétries de la géométrie sous-jacente. S’appuyant sur des travaux fondamentaux antérieurs, la TSRT identifie exactement quatre classes de symétrie causalement admissibles — U(1), SU(2), SU(3) et la courbure gravitationnelle — correspondant chacune de façon unique à l’une des interactions fondamentales connues. Cette classification est à la fois extrêmement restrictive et, dans les limites analysées, empiriquement complète : elle retrouve les groupes de jauge du Modèle Standard et prédit, dans la mesure où aucune autre classe de symétrie tremblante causalement admissible n’a été identifiée, l’absence d’autres forces fondamentales ou générations supplémentaires. Contrairement à la théorie quantique des champs, qui autorise des extensions arbitraires des champs, la TSRT exclut les superpartenaires, les familles additionnelles ou les secteurs de jauge exotiques, à moins qu’ils ne puissent être réalisés comme des déviations métriques causalement cohérentes — une possibilité qui s’avère incompatible avec la structure variationnelle de la théorie dans sa formulation actuelle. La quantification de la charge, la correspondance spin-statistiques, le confinement de la charge de couleur et la discrétisation des familles de particules émergent tous de la cohérence causale et de la clôture topologique des géodésiques tremblantes, sans recours à la quantification des champs ni à une interprétation probabiliste. Au-delà de la classification, la TSRT fournit une fondation géométrique pour des principes traditionnellement considérés comme axiomes. La dérivation de l’énergie de repos et de la masse inertielle découle directement de l’accumulation de l’action au cours du temps propre, donnant la relation d’Einstein entre énergie et masse comme une conséquence déterministe de la résistance géodésique à la courbure. Les interactions sont formulées comme des couplages médiés par la courbure entre les modes tremblants, remplaçant les échanges de particules virtuelles par des corrélations géométriques déterministes. Le boson de Higgs est réinterprété non comme une excitation quantifiée d’un champ scalaire, mais comme une résonance métastable liée à la courbure et se désintégrant par fragmentation géométrique. Enfin, la TSRT fournit des prédictions empiriques qui la distinguent du Modèle Standard, incluant des contraintes géométriques sur les générations de fermions, des échelles d’action minimale liées à la constante de Planck, et des signatures testables du tremblement induit par la courbure. En dérivant l’ensemble complet des particules et interactions connues à partir de premiers principes — tout en identifiant des conditions précises de réfutation — ce travail fait progresser le programme d’unification géométrique et montre que la matière procède d’un principe unique et fondamental : la cohérence causale de l’espace-temps tremblant.Questo articolo dimostra che tutte le particelle fondamentali conosciute emergono come autostati geometrici dello spaziotempo tremante: oscillazioni localizzate e causalmente stabili della metrica dello spaziotempo stesso. Nell’ambito della Teoria della Relatività dello Spaziotempo Tremante (TSRT), l’identità delle particelle, la massa, lo spin e la carica non vengono introdotti come ipotesi indipendenti, ma emergono in modo deterministico come proprietà intrinseche delle deviazioni legate alla simmetria della geometria sottostante. Basandosi su precedenti lavori fondamentali, la TSRT identifica esattamente quattro classi di simmetria causalmente ammissibili — U(1), SU(2), SU(3) e la curvatura gravitazionale — ciascuna corrispondente in modo univoco a una delle interazioni fondamentali note. Questa classificazione è al contempo estremamente restrittiva e, entro i limiti analizzati, empiricamente completa: essa recupera i gruppi di gauge del Modello Standard e predice, nella misura in cui non siano state individuate ulteriori classi di simmetria tremante causalmente ammissibili, l’assenza di altre forze fondamentali o generazioni aggiuntive. In contrasto con la teoria quantistica dei campi, che consente estensioni arbitrarie dei campi, la TSRT esclude superpartner, famiglie aggiuntive o settori di gauge esotici, a meno che non possano essere realizzati come deviazioni metriche causalmente coerenti — un’eventualità che risulta incompatibile con la struttura variazionale della teoria nella sua formulazione attuale. La quantizzazione della carica, la corrispondenza spin-statistica, il confinamento della carica di colore e la discrezione delle famiglie di particelle emergono tutti dalla coerenza causale e dalla chiusura topologica delle geodetiche tremanti, senza ricorrere alla quantizzazione dei campi o a interpretazioni probabilistiche. Oltre alla classificazione, la TSRT fornisce una base geometrica per principi tradizionalmente considerati assiomatici. La derivazione dell’energia di riposo e della massa inerziale deriva direttamente dall’accumulo dell’azione nel tempo proprio, producendo la relazione di Einstein tra energia e massa come risultato deterministico della resistenza geodetica alla curvatura. Le interazioni sono formulate come accoppiamenti mediati dalla curvatura tra i modi tremanti, sostituendo gli scambi di particelle virtuali con correlazioni geometriche deterministiche. Il bosone di Higgs è reinterpretato non come un’eccitazione quantizzata di un campo scalare, ma come una risonanza metastabile legata alla curvatura che decade tramite frammentazione geometrica. Infine, la TSRT fornisce predizioni empiriche che la distinguono dal Modello Standard, tra cui vincoli geometrici sulle generazioni di fermioni, scale di azione minima collegate alla costante di Planck e firme sperimentalmente verificabili del tremolio indotto dalla curvatura. Derivando l’insieme completo delle particelle e delle interazioni conosciute a partire da principi primi — e identificando condizioni precise di falsificazione — questo lavoro fa avanzare il programma di unificazione geometrica e dimostra che la materia origina da un unico principio fondamentale: la coerenza causale nello spaziotempo tremante.В данной работе показано, что все известные фундаментальные частицы возникают как геометрические собственные моды дрожащего пространственно-временного континуума: локализованные, каузально устойчивые колебания самой метрической структуры пространства-времени. В рамках Теории Относительности Дрожащего Пространства-Времени (TSRT) идентичность частиц, их масса, спин и заряд не вводятся в качестве независимых предположений, а детерминированно проявляются как внутренние свойства симметрийно обусловленных отклонений базовой геометрии. Основываясь на предыдущих фундаментальных исследованиях, TSRT выделяет ровно четыре каузально допустимые симметрийные классы — U(1), SU(2), SU(3) и гравитационную кривизну — каждая из которых уникально соответствует одному из известных фундаментальных взаимодействий. Эта классификация одновременно чрезвычайно ограничительна и, в пределах проведённого анализа, эмпирически полна: она воспроизводит калибровочные группы Стандартной модели и предсказывает, поскольку дополнительных каузально допустимых дрожащих симметрийных классов не выявлено, отсутствие новых фундаментальных сил или поколений частиц. В отличие от квантовой теории поля, допускающей произвольные расширения поля, TSRT исключает суперпартнёров, дополнительные семейства или экзотические калибровочные секторы, если только они не могут реализоваться как каузально согласованные метрические отклонения — что в текущей вариационной формулировке теории оказывается невозможным. Квантизация заряда, соответствие спина и статистики, конфайнмент цветового заряда и дискретность семейств частиц — всё это возникает из каузальной согласованности и топологического замыкания дрожащих геодезических, без обращения к квантизации поля или вероятностной интерпретации. Помимо классификации, TSRT предоставляет геометрическое основание для принципов, традиционно считающихся аксиоматичными. Вывод энергии покоя и инерционной массы непосредственно следует из накопления действия по собственному времени, приводя к соотношению Эйнштейна между энергией и массой как детерминированному результату геодезического сопротивления кривизне. Взаимодействия формулируются как опосредованные кривизной связи между дрожащими модами, заменяя обмен виртуальными частицами детерминированными геометрическими корреляциями. Бозон Хиггса переосмысливается не как квантизованное возбуждение скалярного поля, а как метастабильный резонанс, связанный с кривизной и распадающийся через геометрическую фрагментацию. Наконец, TSRT даёт эмпирические предсказания, отличающие её от Стандартной модели, включая геометрические ограничения на поколения фермионов, минимальные масштабы действия, связанные с постоянной Планка, и проверяемые сигналы дрожания, индуцированного кривизной. Выводя полный набор известных частиц и взаимодействий из первых принципов — и указывая точные условия их фальсификации — эта работа продвигает программу геометрического объединения и демонстрирует, что материя происходит из одного фундаментального принципа: каузальной согласованности дрожащего пространства-времени.本文证明，所有已知的基本粒子都是颤动时空的几何本征模式：时空度量本身的局域化、因果稳定的振荡。在颤动时空相对论理论（TSRT）框架下，粒子的身份、质量、自旋和电荷并非作为独立假设引入，而是以确定性的方式，从基础几何中受对称性约束的偏差内在地涌现。基于此前的奠基性研究，TSRT恰好识别出四种因果允许的对称性类别——U(1)、SU(2)、SU(3)以及引力曲率——每一种都唯一对应于一种已知的基本相互作用。这种分类既极为严格，又在已分析的范围内经验上是完备的：它重现了标准模型的规范群，并在尚未发现其他因果允许的颤动对称性类别的前提下，预测不存在更多的基本力或粒子世代。与允许任意场扩展的量子场论不同，TSRT排除了超对称伴侣、额外的家族或奇异的规范部门，除非它们可以表现为因果一致的度量偏差——而这在当前的理论变分结构下被证明是不兼容的。电荷量子化、自旋-统计对应性、色荷禁闭以及粒子家族的离散性，全部源于颤动测地线的因果一致性和拓扑闭合，而无需依赖场量子化或概率性解释。除了分类之外，TSRT还为传统上被视为公理的原理提供了几何基础。静止能量和惯性质量的推导直接来自于固有时间上的作用累积，从而将能量与质量之间的爱因斯坦关系视为测地线对曲率阻抗的确定性结果。相互作用被表述为颤动模式之间由曲率介导的耦合，用确定性的几何相关性取代了虚粒子的交换。希格斯玻色子不再被解释为量化的标量场激发，而是作为一种通过几何破裂衰变的、与曲率绑定的亚稳态共振。最后，TSRT提出了与标准模型不同的经验预测，包括对费米子世代的几何约束、与普朗克常数相关的最小作用尺度，以及曲率诱导颤动的可检验信号。通过从第一原理推导出所有已知的粒子和相互作用，并明确提出可证伪的条件，这项工作推动了几何统一计划，并表明物质源自一个唯一且根本的原理：颤动时空的因果一致性
Mise en œuvre des techniques de traitement du langage naturel pour l'extraction et la formalisation automatique d'exigences dans le domaine de la construction
No full textIn the construction sector, managing technical requirements often relies on unstructured documents, such as the Cahier des Clauses Techniques Particulières (CCTP), a contractual document specifying the requirements that buildings must adhere to. However, manually extracting information from these documents is a lengthy and error-prone process, complicating their integration into automated compliance verification systems. The emergence of GPT-3 in early 2023 marked a turning point in Natural Language Processing (NLP), introducing advanced capabilities in information extraction and text comprehension, challenging traditional methods and the initial expectations of this thesis. The power of GPT-3 and its successor, GPT-4, revealed simpler and more accessible extraction solutions, no longer requiring the same level of technical expertise in NLP as before.In response to these advances, this thesis developed a methodological framework to automate the extraction of technical requirements and evaluate the effectiveness of GPT-4 and other models compared to more traditional methods through a benchmarking process. The methodology includes creating an annotated database of CCTPs for model training and validation, followed by Named Entity Recognition (NER) and Relation Extraction (RE) techniques to identify and link relevant technical entities. Advanced models such as CamemBERT and GPT-4 were adapted to the construction domain through transfer learning and prompt engineering, while more traditional Machine Learning methods, such as the Random Forest (RF) model, were also tested for RE tasks.The results show that after fine-tuning, CamemBERT achieved F1-scores above 96% for NER, while the RF model reached an F1-score of 83% for RE, outperforming GPT-4, which scored 74%. Although GPT-4, using few-shot learning, did not surpass fine-tuned models, its performance remains promising, particularly in achieving competitive results with few examples, opening new perspectives.The extracted entities and relationships are structured for direct integration into compliance verification systems, especially for Building Information Modeling (BIM), allowing for the automatic verification of requirements extracted from CCTPs against digital models. This system ensures more efficient and consistent requirements management, reducing the manual interventions needed to maintain alignment between BIM models and contractual documents.These results open avenues for improving the proposed framework. In particular, creating an extraction system based on a Retrieval-Augmented Generation (RAG) approach could leverage the capabilities of Large Language Models (LLMs), such as GPT, to further improve performance even with limited annotated data. Additionally, the developed framework provides a solution for handling unstructured documents by transforming textual requirements into machine-readable information, facilitating a transition toward Model-Based Requirements Engineering (MBRE). This approach supports smoother collaboration between stakeholders throughout the project lifecycle, enhancing the traceability and consistency of requirements.Dans le domaine de la construction, la gestion des exigences techniques repose souvent sur des documents non structurés, tels que le Cahier des Clauses Techniques Particulières (CCTP), un document contractuel qui spécifie les exigences auxquelles les bâtiments doivent se conformer. Cependant, l’extraction manuelle de ces informations reste un processus long et sujet aux erreurs, rendant difficile leur intégration dans des systèmes automatisés de vérification de conformité. L’émergence de GPT-3 au début de 2023 a marqué un tournant dans le domaine du Traitement Automatique du Langage Naturel (TALN), en introduisant des capacités avancées d’extraction d’information et de compréhension de texte, remettant ainsi en question les méthodes traditionnelles et les attentes de cette thèse. La puissance de GPT-3 et de son successeur GPT-4 a révélé des solutions d'extraction plus simples et plus accessibles, ne nécessitant plus les mêmes compétences techniques en TALN qu’auparavant.En réponse à ces avancées, cette thèse a développé un cadre méthodologique pour automatiser l'extraction des exigences techniques et évaluer l’efficacité de GPT-4 et d’autres modèles par rapport aux méthodes plus classiques, à travers un processus de benchmarking. La méthodologie comprend la création d’une base de données annotée de CCTPs pour l’entraînement et la validation des modèles, suivie de techniques de Named Entity Recognition (NER) et de Relation Extraction (RE) pour identifier et relier les entités techniques. Des modèles avancés tels que CamemBERT et GPT-4 ont été adaptés au domaine de la construction via l’apprentissage par transfert et le Prompt Engineering, tandis que des méthodes de Machine Learning plus classiques, comme le modèle Random Forest (RF), ont également été testées pour la tâche de RE.Les résultats montrent qu’après un fine-tuning, CamemBERT a atteint des F1-scores supérieurs à 96 % pour la NER, tandis que le modèle RF a obtenu un F1-score de 83 % pour la RE, surpassant GPT-4, qui a obtenu un score de 74 %. Bien que GPT-4, en utilisant le Few-shot Learning, n’ait pas surpassé les modèles fine-tunés, ses performances restent intéressantes, en particulier sa capacité à atteindre des résultats compétitifs avec peu d’exemples, ce qui ouvre de nouvelles perspectives.Les entités et relations extraites sont structurées pour une intégration directe dans des systèmes de vérification de conformité, notamment pour le Building Information Modeling (BIM), permettant de vérifier automatiquement les exigences extraites des CCTPs par rapport aux modèles numériques. Ce système assure une gestion des exigences plus efficace et cohérente, réduisant ainsi les interventions manuelles nécessaires pour maintenir la conformité entre le BIM et les documents contractuels.Ces résultats ouvrent des pistes d’amélioration pour le cadre proposé. En particulier, la création d’un système d’extraction basé sur une approche Retrieval-Augmented Generation (RAG) pourrait exploiter les capacités des Large Language Models (LLMs), comme GPT, pour améliorer encore les performances, même avec des données annotées limitées. Le cadre développé offre également une solution pour traiter les documents non structurés en transformant les exigences textuelles en informations exploitables par machine, facilitant ainsi une transition vers une ingénierie des exigences basée sur des modèles (Model-Based Requirements Engineering - MBRE). Cette approche soutient une collaboration plus fluide entre les parties prenantes tout au long du cycle de vie des projets de construction, renforçant la traçabilité et la cohérence des exigences
Experimental investigation of flow field behind a bluff body controlled by an oscillating upstream micro-rod
No full textInternational audienceIn the current study, the flow field due to an aerodynamic interaction between two non-identical cylinders arranged in tandem has been investigated experimentally. The considered system consists of an upstream micro-cylindrical rod oscillating in the transvers direction while the downstream cylindrical bluff body was held stationary. The experiments are carried out at a Reynolds number of 12537 based on the free stream velocity and the main circular cylinder diameter, D = 75.8 mm. The micro-rod is placed in-line to the main cylinder such that the ratio is set to be 2.5 times the formation length of the vortices generated by the small rod. A physical wake analysis is performed to illustrate the mechanism of such active flow control in altering the aerodynamic performance of a circular cylinder. Both the influences of the upstream micro-rod diameter, as well as the oscillation parameters (frequency and amplitude) are deeply examined using a hot-wire anemometry technique. The time mean velocity fields, turbulence intensity and skewness factor are measured for two micro-rod diameters (d/D=2.5% and d/D=5%) and two oscillating amplitudes (A/D=6.6% and A/D=13.2%). Moreover, the vortex shedding frequency of the baseline stationary cylinder is established by hotwire anemometer technique and this frequency is being used to oscillate the upstream micro-cylinder with its sub-harmonic, harmonic and super-harmonic frequency ratios laying in the range of [0 -7]. All these parameters are analysed and their corresponding effects on drag force and turbulence statistics resulted from this flow interference problem. The main results are reported at proper gap spacing which have showed that the oscillating micro-rod significantly contribute to a drag reduction compared to the baseline single cylinder. Furthermore, the wake formation behind the stationary cylinder is influenced by the interaction induced from the oscillatory micro-rod. It is also noticeable; that the oscillating micro-rod is more dominant in drag reduction for the higher diameter as well as the bigger oscillating amplitude especially within the lock-on frequency range. To sum up, placing a micro-rod as an upstream active control in tandem arrangement with larger cylinder has given noteworthy aerodynamic force reduction in term of overall system drag that can reach 35% at optimal oscillating parameters
Experimental study of the influence of the channel geometry of regenerative pumps on their internal flow
No full textInternational audienceRegenerative pumps are capable of generating high pressures at relatively low flow rates. This advantage is particularly useful for electric vehicle cooling systems or for cleaning systems of autonomous vehicle sensors. However, their nominal efficiency is limited. The objective of this study is to better understand the influence of the pump channel geometry on the internal flow of regenerative pumps. A prototype of a regenerative pump with a variable channel geometry was designed and manufactured. The height and the width of the channel are varied but the open channel area is conserved to keep similar flowrates between the different configurations. It is also possible to change the mechanical clearances between the housing and the impeller and between the stripper and the impeller to study their effect on the internal leakages of the pump. Measurements of the hydrodynamic characteristics, including pressure vs. flow rate and efficiency vs flowrate, will be conducted. This study will provide a better understanding of the performance control parameters of this type of machine. These experimental results will be compared to analytical results from 1D models of the literature
The structure of liquid carbon elucidated by in situ X-ray diffraction
No full textInternational audienceCarbon has a central role in biology and organic chemistry, and its solid allotropes provide the basis of much of our modern technology1. However, the liquid form of carbon remains nearly uncharted2, and the structure of liquid carbon and most of its physical properties are essentially unknown3. But liquid carbon is relevant for modelling planetary interiors4,5 and the atmospheres of white dwarfs6, as an intermediate state for the synthesis of advanced carbon materials7,8, inertial confinement fusion implosions9, hypervelocity impact events on carbon materials10 and our general understanding of structured fluids at extreme conditions11. Here we present a precise structure measurement of liquid carbon at pressures of around 1 million atmospheres obtained by in situ X-ray diffraction at an X-ray free-electron laser. Our results show a complex fluid with transient bonding and approximately four nearest neighbours on average, in agreement with quantum molecular dynamics simulations. The obtained data substantiate the understanding of the liquid state of one of the most abundant elements in the universe and can test models of the melting line. The demonstrated experimental abilities open the path to performing similar studies of the structure of liquids composed of light elements at extreme conditions