Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Efficient Implementation of Neural Networks Usual Layers on Fixed-Point Architectures
No full textInternational audienceConvolutional Neural Network (CNN) have been widely used in image classi-cation. Over the years, they have also beneted from various enhancements andthey are now considered state-of-the-art techniques for image-like data. However, when they are used for regression to estimate some function value fromimages, few recommendations are available to construct robust CNN regressormodels. In this study, a robustness enforcing mechanism is proposed for CNNregression models. It combines convolutional neural layers to extract high levelfeatures representations from images with a soft labelling technique that helpsgeneralization performance. More specically, as the deep regression task ischallenging, the idea is to account for some uncertainty in the targets that areseen as distributions around their mean. Building from earlier work (Imani &White, 2018), a specic histogram loss function based on the Kullback-Leibler(KL) divergence is applied during training. The prior distributions are selectedaccording to the physical characteristics of the parameters to estimate. To assess and illustrate the technique, the model is applied to Global NavigationSatellite System (GNSS) multipath estimation where multipath signal parameters have to be estimated from correlator output images from the I and Qchannels. The multipath signal delay, magnitude, Doppler shift frequency andphase parameters are estimated from synthetically generated datasets of satellite signals. Experiments are conducted under various receiving conditions andvarious input images resolutions to test the estimation performances qualityand robustness. The results show that the proposed soft labelling CNN technique using distributional loss outperforms classical CNN regression under allconditions. Furthermore, the extra learning performance achieved by the mode
Energy consumption of Aircraft with new propulsion systems and storage media
No full textInternational audienceThe transition from fossil fuels to the use of sustainable energy sources is an importantchallenge for aviation. A particularly critical question is to identify the best mix of energiesfor the future air transport system. In this quest, the capability to simulate various complexscenarios, varying assumptions, models, and objectives, will play an important part. Thispaper provides a simple and very rapid tool to evaluate the energy consumption of parametricairplane configurations with thermal propulsion powered by kerosene, methane, hydrogen, orelectric propulsion powered by hydrogen fuel cells or batteries. The maturity of the differenttechnologies is driven by a small number of global indices that are easy to connect with stateof the art or forecasts. The user can freely select the capacity, range, speed, and propulsionsystem of the airplane, and simulate its operations over any missions within its payload-rangeenvelope. The tool has been validated on its capability to reproduce the characteristic weightsof a range of very different aircraft, from A380 to general aviation airplanes. Finally, threeuse cases are presented. The first one allowed us to identify the level of technological challengeof the Alice aircraft project from EVIATION. The second one is a short mapping of thetransportation potential of five different propulsion systems for the segment of 19-passengercommuter airplanes. The third use case puts the thermal hydrogen aircraft in the perspective ofexisting kerosene airplanes in terms of energy and structural efficiency, and reveals an optimumof the hydrogen technology
A Preliminary Study on the Predictive Validity of Neural Efficiency for the Air Traffic Controllers Selection
No full textInternational audienc
Towards More Automated Airport Ground Operations Including Engine-Off Taxiing Techniques Within the Auto-Steer Taxi at AIRport (ASTAIR) Project
No full textInternational audienceThis paper discusses SESAR's Auto-Steer Taxi at Airport (ASTAIR) project, which seeks to advance airport ground operations including engine-off taxiing to move towards sustainable airports. The ASTAIR concept integrates human-AI teaming to optimize aircraft movement from gates to runways, with the primary objectives of improving predictability, efficiency, and environmental sustainability at large airports. Building on previous initiatives such as SESAR's AEON, ASTAIR brings high-level automation to tasks like autonomous taxiing and vehicle routing. The system assists operators by calculating conflict-free routes for vehicles and dynamically adjusting operations based on real-time data. Based on workshops with several stakeholders, we describe the operational challenges involved in implementing ASTAIR, including managing parking stand availability and adapting to unforeseen events. A significant challenge highlighted is the human-automation partnership, where AI plays a supportive role but humans retain control over critical decisions, particularly in cases of system failure. The need for clear and consistent collaboration between AI and human operators is emphasized to ensure safety, efficiency, and improved compliance with take-off schedules, which in turn facilitates in-flight optimization
Mixed-integer quadratic programming formulations for computing the Lipschitz constant of ReLU networks
No full textInternational audienceThe Lipschitz constant of a neural network is a useful metric to get information about the robustness of a trained network. Its exact calculation is however NP-hard even for one hidden layer ReLU networks. In this presentation, by taking into account activation regions at each layer as new constraints, we propose new quadratically constrained mixed-integer programming formulations for the neural network Lipschitz estimation problem. The solutions of these problems give lower bounds and/or upper bounds of the Lipschitzconstant. We give some experiments to compare the proposed approach with State Of the Art algorithms to estimate the Lipschitz constant of a neural network
Optimisation multidisciplinaire intégrant la contrôlabilité pour la conception de drones à décollage et atterrissage verticaux.
No full textIn the last decades, the continuous improvement in areas such as on-board electronics, flight control,and additive manufacturing has allowed the field of aerial robotics to evolve considerably. Theso-called drones are now part of our everyday life, commonly employed for commercial, recreational,and military purposes. Multicopter vehicles are arguably the most widespread, mostly due to theirsimplicity in operations, which makes them suitable for applications that demand flight dexterity,including video recording, manipulation, and human-robot iteration tasks. However, they oftenhave limited payload capacity, flight range and endurance. Differently, winged vehicles are bettersuited for longer flights and payload capacity, but are more challenging to control and generallytakeoff and land horizontally, necessitating a dedicated runway. In this context, the class of hybridvehicles, characterized by the presence of wings and the capability of taking off and landing vertically,has been attracting the attention of the aerial robotics community. “VTOL” (vertical takeoffand landing) robots represent a middle ground solution, capable of achieving long distances withpayload, while maintaining a simplified operation. Their application in real life scenarios is stillscarce, and there are technical challenges that need to be solved to allow their widespread application.The main contribution of this work is in the formulation of a vehicle design methodologyfor VTOL aerial robots. The thesis starts by identifying the main gaps related to vehicle design,disciplinary modeling, and control. Then, it presents the initial definition of a multidisciplinarydesign and optimization based design procedure. This methodology is used to conceive a vehiclefor a drone competition, which is then evaluated with wind tunnel and flight test campaigns andredesigned. The second version of the vehicle is then assessed in competition and through furtherflight tests, demonstrating endurance close to the desired level. With this experience, the need to account for more disciplines, notably propeller performance and flight dynamics, is validated.Such studies are also presented and gradually added to the optimization problem. Various existingmethods for evaluating propeller performance in oblique flow are studied and compared with experimentalresults, allowing the most suitable to be selected and integrated into the design process.A semi-automated method for 3D vehicle model generation is then presented, improving weightand inertia prediction while reducing the time between design and test flights. In order to increasethe vehicle’s ability to reject disturbances “by design”, a closed loop flight dynamics analysis isadded to the optimization problem. With this new component, the path error in windy conditionsis reduced, and stays within the defined bounds. Lastly, the notion of mission adaptability is alsoadded, and an optimization problem that couples altogether design, control, and trajectory optimizationis presented. Such an integrated design process, as well as the study regarding disciplinaryanalysis and vehicle manufacturing, could be helpful for the automatic design of mission tailoredVTOL robots.Au cours des dernières décennies, le domaine de la robotique aérienne a considérablement évolué.Les améliorations constantes dans l’électronique embarquée, les commandes de vol et la fabricationadditive ont permis la prolifération de petits véhicules volants. Les « drones » font désormaispartie de notre vie quotidienne et sont couramment utilisés à des fins commerciales, récréatives etmilitaires. Les multicoptères sont les plus répandus en raison de leur simplicité d’utilisation. Ils conviennentaux applications nécessitant une grande dextérité en vol, comme l’enregistrement vidéo, lamanipulation et les tâches d’itération homme-robot. Cependant, ils présentent souvent une capacitéde charge, un rayon d’action et une endurance limités. Les véhicules à voilure fixe sont mieuxadaptés aux vols de longue durée, car leur poids est équilibré par la portance générée par les ailes,améliorant ainsi l’efficacité du vol et la capacité de charge utile. Cependant, ils sont plus difficilesà contrôler et nécessitent une piste d’atterrissage spécifique. Les véhicules hybrides, caractériséspar la présence d’ailes et la capacité de décoller et d’atterrir verticalement, ont attiré l’attention dela communauté de la robotique aérienne. Les robots à décollage et atterrissage verticaux (VTOL)représentent une solution intermédiaire, capable de parcourir de longues distances avec une chargeutile tout en conservant un fonctionnement simplifié. Leur application dans des scénarios réels estencore rare, et des défis techniques doivent être résolus pour permettre leur application à grandeéchelle. La principale contribution de ce travail est la formulation d’une méthodologie de conceptionde véhicules pour les robots aériens VTOL. La thèse commence par identifier les principaleslacunes liées à la conception du véhicule, à la modélisation disciplinaire et au contrôle. Ensuite,elle présente une procédure de conception multidisciplinaire et d’optimisation basée sur trois disciplines: l’aérodynamique, la masse, et la mission. Cette méthodologie est utilisée pour concevoir un véhicule pour une compétition de drones, testé avec des campagnes d’essais en soufflerie et envol, entraînant des modifications dans la conception. La deuxième version du véhicule est évaluéeen compétition et par d’autres essais en vol, démontrant une endurance proche du niveau souhaité.Grâce à cette expérience, la nécessité de prendre en compte d’autres disciplines, notamment lesperformances de l’hélice et la dynamique du vol, est validée. Ces études sont progressivementajoutées au problème d’optimisation. Diverses méthodes existantes pour évaluer les performancesdes hélices dans un écoulement oblique sont étudiées et comparées aux résultats expérimentaux,permettant de sélectionner la plus appropriée et de l’intégrer dans le processus de conception. Uneméthode semi-automatique pour la génération de modèles de véhicules en 3D est présentée, améliorantla prédiction de la masse et de l’inertie tout en réduisant le temps entre la conception et lesessais en vol. Les drones sont sujets aux perturbations atmosphériques, et le manque de robustessedu premier véhicule conçu confirme cette hypothèse. Pour résoudre ce problème, une analyse dela dynamique de vol en boucle fermée est ajoutée au problème d’optimisation. Cet ajout permetd’utiliser l’erreur de suivi de trajectoire comme contrainte, conduisant le processus de conceptionvers un véhicule plus robuste. Grâce à cette nouvelle composante, l’erreur de trajectoire en casde vent est réduite et reste dans les limites définies. Enfin, la mission peut être modifiée et unproblème d’optimisation qui associe à la fois la conception, le contrôle et la trajectoire est présenté.Un tel processus de conception intégré, ainsi que l’étude concernant l’analyse disciplinaire et lafabrication du véhicule, pourraient être utiles pour la conception automatique de robots VTOLadaptés à la missio
Impact of Temporal Correlation of Errors on GPS Receiver Autonomous Integrity Monitoring
No full textInternational audienceReceiver Autonomous Integrity Monitoring (RAIM) is widely adopted in commercial aircraft to support integrity monitoring for aircraft navigation safety in en-route and Non-Precision Approach (NPA) phases. Recently, Advanced RAIM (ARAIM) has been developed, which further considers the simultaneous satellite failures in multi-constellation systems and supports multi-frequency signals. Standards provide a test procedure for onboard RAIM, offering a means to demonstrate compliance with integrity risk requirements for civil aviation. Based on this requirement, the aircraft computes the protection level that bounds the position error. Unlike the integrity risk requirement, which is defined over the entire operational period, the protection level is computed on a per sample or per-epoch basis. To properly interpret the requirement defined over the operation period to a single time instance, the temporal correlation of error sources within the operational period must be considered and quantified by the number of effective samples (NES). While recent work has evaluated NES for ARAIM, the NES for RAIM has not been thoroughly investigated. Based on the test procedure recommended in the Minimum Operational Performance Standards (MOPS), this paper focuses on estimating and bounding the NES for RAIM under different scenarios of measurement errors. Moreover, to address the limitations of these test procedures in light of the results obtained, representative test procedures are proposed to account for worst-case fault bias and address the impact of temporal correlation
Mental incapacitation makes pilot react later, miss information, take poor decisions and crash more often: Mental incapacitation in aviation: a scoping review
No full textInternational audienceIn aviation, mental incapacitation, also known as cognitive incapacitation, refers to a temporary decline in pilot cognitive abilities due to stressful events, leading to performance degradation and accidents. While medical incapacitation is well-studied, understanding of mental incapacitation lacks systematic review articles. A scoping review following PRISMA guidelines was conducted and included 28 relevant articles out of 935 after a dual screening procedure. Focus was mainly on experimental research, with 74% investigating unexpected, surprising and startling events. Most studies (70%) utilised flight simulators, inducing incapacitation via in-flight situations like system failures, weather conditions and misleading automation system behaviours. The most frequent conceptual background chosen was the Reframing Model from Landman et al. (2017). The accumulated population in our review was of 546 pilots with an average age of 34.9yo (sd = 8.12yo). Data collection usually involved physiological data (44% of the articles), flight data (55% of the articles), and cognitive/behavioural performance (66% of the articles). Results showed increased pupil dilation, delays and negative impact on time spent looking at relevant areas. Increased galvanic skin response, and heart rate, along with EEG evidence of brain activity changes in fronto-central and parietal regions. Cognitive effects included inattentional deafness, increase in response times, delayed decisions, channelised attention, loss of accuracy, increase in selfreported workload, missed detection of mode change, poor understanding of issues, delays in check-list completion and missed communications. Mental incapacitation impacted flight parameters, causing altitude loss, delayed go-around, poor throttle management, automation resetting perseveration and evencrashes. Studies predominantly examined unexpected events and high workload situations; exploring other contexts like crew conflicts and miscommunication is warranted. Because of the variety of phenomenons related to mental incapacitation, existing studies lack accuracy in assessing mental incapacitation intensity and replicability. Future research should investigate additional sources of incapacitation and also enhance understanding of its effects by reducing uncertainty and precising underlying factors favouring its emergence
Experimental UAV Flights to Collect Data Within Cumulus Clouds
No full textInternational audienceThis article presents the deployment of micro UAVs to study the evolution of cumulus clouds. The dynamic nature of the environment, the difficult weather conditions, the long distances, and the limited flight performances of micro UAV systems make such missions quite challenging. After describing the missions constraints and objectives, the system's main component is depicted: it is the definition of adaptive flight patterns that allow the UAVs to track the areas of interest in the clouds autonomously, using real-time sensor readings. The system architecture is then presented, considering the information feedback and decision process made by the operators and scientists on the ground and the necessary flight autonomy of the UAVs. The complete system has been deployed during an international meteorological campaign on Barbados island, and extended with extra test flights afterward. Details of the operational organization and the achieved flights are reported. The lessons learned during the field campaign revealed the strength and weaknesses of the proposed system and possible improvements are discussed. The collected data has contributed to a better understanding of cloud evolution, demonstrating that a tight coupling of the sensors and the flight control system is a crucial point for extending the performances of UAV systems in atmospheric science