Portail HAL ONERA
Not a member yet
12842 research outputs found
Sort by
SMT-EX: An Explainable Surrogate Modeling Toolbox for Mixed-Variables Design Exploration
No full textInternational audienceSurrogate models are of high interest for many engineering applications, serving as cheap-toevaluate time-efficient approximations of black-box functions to help engineers and practitioners make decisions and understand complex systems. As such, the need for explainability methods is rising and many studies have been performed to facilitate knowledge discovery from surrogate models. To respond to these enquiries, this paper introduces SMT-EX, an enhancement of the open-source Python Surrogate Modeling Toolbox (SMT) that integrates explainability techniques into a state-of-the-art surrogate modelling framework. More precisely, SMT-EX includes three key explainability methods: Shapley Additive Explanations, Partial Dependence Plot, and Individual Conditional Expectations. A peculiar explainability dependency of SMT has been developed for such purpose that can be easily activated once the surrogate model is built, offering a user-friendly and efficient tool for swift insight extraction. The effectiveness of SMT-EX is showcased through two test cases. The first case is a 10-variable wing weight problem with purely continuous variables and the second one is a 3-variable mixed-categorical cantilever beam bending problem. Relying on SMT-EX analyses for these problems, we demonstrate its versatility in addressing a diverse range of problem characteristics. SMT-Explainability is freely available on Github.Les modèles de substitution suscitent un grand intérêt dans de nombreuses applications en ingénierie, car ils offrent des approximations rapides et peu coûteuses de fonctions complexes en "boîte noire". Ces modèles aident les ingénieurs et les praticiens à prendre des décisions et à mieux comprendre les systèmes complexes. Ainsi, le besoin de méthodes d'explicabilité ne cesse de croître, et de nombreuses études ont été menées pour faciliter l'extraction de connaissances à partir de modèles de substitution. Pour répondre à ces besoins, cet article présente SMT-EX, une amélioration de la boîte à outils open-source Python Surrogate Modeling Toolbox (SMT), qui intègre des techniques d'explicabilité dans un cadre de modélisation de substitution à la pointe de la technologie.Plus précisément, SMT-EX inclut trois méthodes clés d'explicabilité : les explications additives de Shapley (SHAP), les graphes de dépendance partielle (PDP) et les espérances conditionnelles individuelles (ICE). Une dépendance d'explicabilité spécifique à SMT a été développée à cet effet et peut être facilement activée une fois le modèle de substitution construit, offrant ainsi un outil intuitif et performant pour une extraction rapide des informations.L'efficacité de SMT-EX est illustrée à travers deux cas tests. Le premier cas est un problème de poids d'aile avec 10 variables purement continues, tandis que le second concerne un problème de flexion d'une poutre en porte-à-faux avec 3 variables mixtes catégoriques. En s'appuyant sur les analyses réalisées avec SMT-EX pour ces problèmes, nous démontrons sa polyvalence face à une grande diversité de caractéristiques de problèmes. SMT-EXplainability est disponible gratuitement sur GitHub
Techniques multi-dimensionnelles de thermométrie par luminescence de composés inorganiques pour la recherche en mécanique des fluides et énergétique
No full textTemperature measurements are key to gain insight into thermochemical processes. They are also central to a wide range of heat transfer or aerodynamics studies. Beyond absolute temperature measurements, extraction of heat transfer rates, or coupled fluctuations terms requires techniques with high spatial and temporal resolution and the simultaneous measurement of multiple quantities such as joint temperature and velocity measurements. From 2010 to 2025, this research has centered around the development and application of luminescence diagnostics. Luminescence thermometry in particular exploits light converting temperature-sensitive material, which emission can be used for remote thermometry. In this thesis, these activities are reviewed and future research directions are also proposed. This thesis is articulated along three lines of activities: the development of imaging techniques, the development of phosphor materials, and the applications of these imaging techniques and phosphor materials to fluidmechanics research.Les mesures de température sont essentielles pour mieux comprendre certains processus thermochimiques clé. Elles sont aussi au cœur d'un large éventail d'études sur le transfert de chaleur ou l'aérodynamique. Au-delà des mesures de la température absolue, l'extraction des taux de transfert de chaleur ou des termes de fluctuations couplées nécessite des techniques à haute résolution spatiale et temporelle et la mesure simultanée de plusieurs grandeurs telles que la température et la vitesse. De 2010 à 2025, cette recherche s'est concentrée sur le développement et l'application des diagnostics par luminescence. La thermométrie par luminescence exploite en particulier des matériaux convertisseur de lumière sensibles à la température, dont l'émission peut être utilisée pour la thermométrie à distance. Dans cette thèse, ces activités sont passées en revue et l’orientation des activités à venir est discutée. Cette thèse s'articule autour de trois axes d'activités : le développement de techniques d'imagerie, le développement de matériaux luminescents sensibles à la température et les applications de ces techniques d'imagerie et de ces matériaux luminescents à la recherche en mécanique des fluides
Sélection de maillage basée sur les erreurs pour des simulations numériques efficaces en présence de paramètres variables
No full textAdvanced numerical simulations often depend on mesh refinement techniques to manage discretization errors in complex models and reduce computational costs. This work concentrates on Adaptive Mesh Refinement (AMR) for steady-state solutions, which uses error estimators to iteratively refine the mesh locally and gradually tailor it to the solution. AMR requires evaluating the solution across a series of meshes. When solving the model for multiple operating conditions, such as in uncertainty quantification studies, full systematic adaptation can cause significant computational overhead. To mitigate this, the Error-based Mesh Selection (EMS) method is introduced to decrease the cost of adaptation. For each operating condition, EMS seeks to choose, from a library of pre-adapted meshes, the one that minimizes the discretization error. A key feature of this approach is the use of Gaussian Process models to predict the solution errors for each mesh in the library. These error models are built solely from the library's meshes and their solutions, using restriction errors as proxies for discretization errors, thereby avoiding additional model evaluations. The EMS method is tested on an analytical shock problem and a supersonic scramjet configuration, showing near-optimal mesh selection. The influence of library size on the resulting error level is also examined.Les simulations numériques avancées s’appuient souvent sur des techniques de raffinement de maillage pour contrôler les erreurs de discrétisation de modèles complexes et réduire les coûts de calcul. Ce travail porte sur le raffinement adaptatif de maillage (AMR) pour des solutions stationnaires, utilisant des estimateurs d’erreur pour raffiner itérativement le maillage localement et l’adapter progressivement à la solution. L’AMR nécessite d’évaluer la solution sur une série de maillages. Lorsqu’on résout le modèle pour de multiples conditions de fonctionnement, comme dans des études de quantification d’incertitudes, une adaptation systématique complète peut entraîner un surcoût de calcul important. Pour y remédier, la méthode de sélection de maillage basée sur l’erreur (EMS) est introduite afin de réduire le coût de l’adaptation. Pour chaque condition de fonctionnement, EMS cherche à choisir, dans une bibliothèque de maillages pré-adaptés, celui qui minimise l’erreur de discrétisation. Une caractéristique clé de cette approche est l’utilisation de modèles à processus gaussiens pour prédire les erreurs de solution pour chaque maillage de la bibliothèque. Ces modèles d’erreur sont construits uniquement à partir des maillages de la bibliothèque et de leurs solutions, en utilisant les erreurs de restriction comme substituts des erreurs de discrétisation, évitant ainsi toute évaluation supplémentaire du modèle. La méthode EMS est testée sur un problème analytique de choc et sur une configuration de statoréacteur supersonique (scramjet), montrant une sélection de maillage quasi optimale. L’influence de la taille de la bibliothèque sur le niveau d’erreur obtenu est également examinée
Cetane Number and Lubricity of Jet Fuel Blends with Alternative Fuels: Additive Effects and Predictive Models
No full textInternational audienceThe integration of synthetic blend components (SBCs) into conventional jet fuels represents a key strategy for decarbonizing the aviation sector. However, under NATO’s single fuel policy, which promotes the use of F-34 jet fuel for both aeronautical and land-based diesel applications, the introduction of SBCs raises concerns regarding their compatibility with diesel engines, particularly in terms of autoignition and lubricity. This study investigates the autoignition and lubricity performance of three fossil jet fuels, representative of global aromatic and sulfur variability, and four SBCs: Alcohol to Jet (ATJ), Hydroprocessed Esters and Fatty Acids (HEFA), HEFA with aromatics (HEFA-A), and Catalytic Hydrothermolysis Jet (CHJ), tested neat and in blends at 30, 50, and 70%v/v fractions. The effect of the S-1750 additive on cetane number (CN) and lubricity was also evaluated at concentrations of 1000, 2000, and 4000 ppmv. Results confirm that CN varies strongly with the type and proportions of SBC, with a notable reduction in CN and additive effectiveness as the ATJ content increases. For ATJ, to achieve the ignition quality of a neat conventional jet fuel, 4000 ppmv of S-1750 is required. Diesel engines would require even higher CN, implying that either the SBC or the additive should be revised. Lubricity measurements using the HFRR method indicated that base fuels and blends exceed the diesel limit without additives, while the BOCLE method confirmed compliance with aviation requirements. The comparison of both methods reveals that the HFRR method appears to be less sensitive to detecting differences among SBC fuels, making it less appropriate for SBC lubricity assessment. In this context, modeling emerges as a complementary approach to enhance understanding and evaluation of SBC fuels and their blends, thereby helping to ensure the safe use of the fuel. A linear mixing rule based on fuel composition predicted a CN within the experimental uncertainty (∼3 CN units), and an empirical model based on physicochemical properties achieved even higher accuracy (R2: 0.984). While the results are promising and the models provide useful foundations for preliminary analysis, the data set should be expanded to support precise and reliable predictions. A linear regression model for HFRR lubricity showed a limited predictive capability (adjusted R2: 0.53), suggesting the need for expanded data sets and advanced nonlinear approaches. Moreover, the reliability of the method used may be reconsidered, as its low sensitivity to detect lubricity differences in SBC blends could explain the poor prediction for these blends. Lastly, several models were tested for predicting the CN boost from the S-1750 additive. Only the optimized version of the model proposed by Thompson, A. Prediction And Precision Of Cetane Number Improver Response Equations SAE Technical Paper 972901, 1997; pp. 8. https://www.sae.org/papers/prediction-precision-cetane-number-improver-response-equations-972901. (accessed 07 October 2025). and a multiple linear regression (MLR) approach achieved strong performances on the test set (R2: 0.97 and 0.93, respectively); however, both struggled to capture the zero boost
Optimizing Space Surveillance Capabilities: System Analysis and Competitive Procurement for PROVIDENCE's 2.5m Telescope
No full textInternational audienceONERA – The French Aerospace Laboratory has launched the PROVIDENCE project to develop a 2.5m-class telescope for Space Domain Awareness (SDA) by the end of 2028. The telescope, optimized for High Angular Resolution with Adaptive Optics, will enable detection, identification, and monitoring of LEO and GEO objects. We expect this unique Optical Ground Station in Europe to trigger international collaboration. We discuss the preliminary design of the telescope and the expected performance of the Adaptive Optics (AO) system that will be dedicated to LEO satellite (spectral)-imaging
Texturation 3D par laser femtoseconde d'optiques à gradient d'indice pour l'optronique infrarouge
No full textInternational audienceThe Size, Weight, and Power (SWaP) challenge drives the development of compact, lightweight, and energy-efficient optical systems, particularly for infrared imaging and multispectral optronics. Femtosecond-Laser- Direct-Writing (FLDW) enables permanent, three-dimensional modification of the refractive index inside transparent materials. This opens new degrees of freedom for implementing gradient-index (GRIN) and freeform optics in bulk substrates. In this work, we investigate the femtosecond laser inscription of optical functions in Corning 9754 commercial glass, transparent from visible to mid-IR. We characterize the processing windows that yields homogeneous, low-scattering refractive index modifications, and we develop a quantitative link between laser parameters and induced optical phase. These results enable the design and fabrication of GRIN microlenses and phase components, evaluated through both wavefront measurements and imaging performances. The proposed workflow demonstrates the potential of femtosecond laser inscription for the realization of optical functions, paving the way for compact architectures dedicated to SWaP-limited applications
Electromagnetic Characterization in a Thermally Constrained Environment at ONERA
No full textInternational audienceThis paper presents the electromagnetic characterization of materials and components under thermal constraints conducted at the Electromagnetism and Radar Department (DEMR) of ONERA. Two test benches are currently under development. The first is a tubular furnace used to evaluate RF sub-assemblies at varying temperatures. The second test bench is based on reflection measurements conducted in free space and employs the CONSERT system and an infrared radiator
Approches basées sur le dialogue pour une reconnaissance d'images transparente
No full textDespite their impressive performance, modern foundation models for image recognition still struggle with complex reasoning involving multiple sources of information and can exhibit unexpected behaviours such as spurious correlations or inaccurate predictions. Moreover, their decision processes remain opaque, limiting trust and interpretability. This thesis introduces dialogue-based image recognition to address these challenges, a novel paradigm for performing and understanding visual recognition through structured interactions between agents. Three frameworks have been developed. The first, inspired by argumentative dialogue, enables two agents to transparently deliberate over image classification using global prototype similarities and local visual attributes derived from foundation models. The second extends the dialogue-based approach to complex scene understanding, where agents combine visual information and a knowledge base to verify image descriptions. The third explores human-machine dialogue for model editing, allowing users to identify and correct unexpected behaviours. Together, these contributions establish the first exploration of dialogue as a foundation for explainable visual recognition, showing how it can tackle key challenges of opacity, complex reasoning, and unexpected behaviours in image recognition.Malgré leurs performances impressionnantes, les modèles de fondation modernes pour la reconnaissance d'images ont encore du mal à gérer les raisonnements complexes impliquant plusieurs sources d'informations et peuvent présenter des comportements inattendus tels que des corrélations fallacieuses ou des prédictions inexactes. De plus, leurs processus décisionnels restent opaques, ce qui limite leur confiance et leur interprétabilité. Cette thèse présente la reconnaissance d'images basée sur le dialogue pour relever ces défis, un nouveau paradigme permettant d'effectuer et de comprendre la reconnaissance visuelle grâce à des interactions structurées entre agents. Trois approches ont été développées. La première, inspirée du dialogue argumentatif, permet à deux agents de délibérer de manière transparente sur la classification d'images à l'aide de similarités globales entre images et d'attributs visuels locaux dérivés de modèles de fondation. La deuxième étend l'approche basée sur le dialogue à la compréhension de scènes complexes, où les agents combinent des informations visuelles et et une base de connaissances pour vérifier les descriptions d'images. La troisième explore un dialogue homme-machine pour l'édition de modèles, permettant aux utilisateurs d'identifier et de corriger les comportements inattendus d'un modèle. Ensemble, ces contributions constituent la première exploration du dialogue comme fondement d'une reconnaissance visuelle explicable, montrant comment il peut relever les défis clés de l'opacité, du raisonnement complexe et des comportements inattendus en reconnaissance d'images
Co-design Robots and Structures Framework for Automated Construction of Modular Space Platforms
No full textInternational audienceLarge-scale space structures have taken center stage as the future of space exploration, paving the way for ambitious endeavors like sprawling solar power stations, intricate telescopes, and giant space stations. This program lies not in manual assembly, but in empowering space robots to autonomously build these structures using plans provided by dedicated on-orbit assembly planning algorithms.In this paper, we describe an approach to large-scale space structure assembly frameworks, keeping in mind the many benefits associated with the use of autonomous crawling mobile robots.Our research presents a model of a large-scale deployable structure, along with its physical constraints, and a corresponding assembly approach. Initially, all building elements (beams and nodes) are stored in the launcher fairing. These will be deployed to form the structure on which the robots themselves will evolve until the final configuration is reached.The proposed concept considers a structure made up of truss beams with standard interfaces at their ends and attachment nodes, also equipped with standard interfaces, supporting a payload such as a solar panel or a deployable antenna. Assembly is performed by crawling robots, which are autonomous systems that adhere to or grip the structure and move around it. However, the robot's plan, spacecraft's actuators, and payload structure should be carefully co-designed to ensure manageable and stable dynamics.Therefore, we represent the assembling problem as an automated planning instance, where several structural constraints dictate the actions available for execution, and their application is constrained in time to avoid introducing dynamic forces that compromise stability and pointing accuracy.This model and its associated tools help to illustrate the feasibility of the approach and the benefits of having AI tools for autonomous assembly robots in space
