Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Détection automatique multi-échelle et de grande envergure d'oscillations intracérébrales pathologiques dans l'épilepsie par réseaux de neurones artificiels
No full textAlmost a third of epileptic patients are resistant to medication. The only way to cure them is to remove the area of the brain that causes the seizures, called the epileptogenic zone (EZ). To locate this area, it is sometimes necessary to carry out stereo-electroencephalography (SEEG) investigations. SEEG consists of implanting intracerebral electrodes in the patient, who remains in hospital for about ten to fifteen days. During this period, the patient's intracerebral activity is continuously recorded on more than a hundred recording channels distributed in the brain structures suspected of being involved in the epileptogenic network. The analysis of the EEG signal by neurologists is a crucial step in the diagnosis, but the amount of data generated is tremendous. As a result, only a small fraction of the recordings can be analyzed by medical teams, who focus mainly on activity during and immediately surrounding seizures. To characterise the extent and dynamics of the epileptogenic network, neurologists also study interictal markers. But some of these biomarkers are impossible to detect manually. The first objective of this interdisciplinary thesis work was to establish new methods to efficiently and automatically detect intercritical markers, namely interictal epileptic discharges (IEDs) and in particular fast ripples (FRs). The second objective was to define and describe the interest of recording pathophysiological markers of epilepsy using micro-electrodes, whereas most studies until now used classical macro-electrodes. Finally, the third objective was focused on FRs, with the idea to better understand their origin, emergence and involvement in the pathology. Our teams use hybrid electrodes that allow for a multi-scale recording of the brain signal of patients. These electrodes are made up of macro-channels allowing the recording of the activity of large neuronal populations and micro-channels capable of capturing much more focal signals, down to the scale of single neuron activity. We have built an automatic IED detector based on a new method of processing the image-transformed signal using a technique we call Convolutional Kernel Density Estimation (CKDE). We also developed an automatic FR detector based on a three-step, CNN-based, ecological approach, mimicking the work of the neurologist. All these tools have been incorporated into graphical user interfaces (GUIs) that combine the different functionalities for easy and efficient use. The detection of IEDs by CKDE offers proof of concept that a pixel-oriented analysis of EEG activity can be used as a strategy to detect interictal markers. We evaluated this method on 10 minutes of recordings in a patient. Fifteen IEDs were automatically detected, of which 13 were true positives and 2 false positives. However, our main results concern the detection of FRs, which would have the greatest potential in the diagnosis of drug-resistant epilepsies. To train the CNN, which is a key component of our detector, we built a database of 4,954 manually detected FRs in 13 patients at both the EEG-macro and the EEG-micro scales. This multi-scale FR detector was incorporated into the software we designed, called Ladybird, which was used in 29 patients to detect and treat several thousand FRs. The technical and theoretical advances made during this thesis allow us to consider a large-scale use of our tools. Our objective is that medical teams can benefit directly from them, in their diagnostic routine. A patent has been filed in view of an industrialization process.Environ un tiers des patients épileptiques sont résistants aux médicaments. La seule solution pour les guérir est de retirer la zone cérébrale à l'origine des crises, appelée zone épileptogène (ZE). Pour localiser cette zone, il est parfois nécessaire des mener des explorations par stéréo-électroencéphalographie (SEEG). L'analyse du signal EEG par les neurologues est une étape déterminante du diagnostic, mais la quantité de données générée est colossale. Ainsi, seule une petite partie des enregistrements peut être analysée par les équipes médicales qui se concentrent principalement sur l'activité durant les crises et celle juste autour. Pour caractériser l'étendue et la dynamique du réseau épileptogène, les neurologues étudient aussi des marqueurs intercritiques. Mais certains de ces biomarqueurs sont strictement invisibles à l'œil nu. Le premier objectif de ce travail de thèse interdisciplinaire consistait à établir de nouvelles méthodes pour détecter efficacement et automatiquement les marqueurs intercritiques, à savoir les pointes épileptiques intercritiques (PEIs) et en particulier les fast ripples (FRs). Le second objectif visait à définir et décrire l'intérêt d'enregistrements des marqueurs physiopathologiques de l'épilepsie par l'intermédiaire de micro-électrodes, alors que la plupart des études jusqu'à présent utilisaient des macro-électrodes classiques. Enfin, le troisième objectif était focalisé sur les FRs, avec pour idée de mieux comprendre leur origine, leur émergence et leur implication dans la pathologie. Nos équipes utilisent des électrodes hybrides permettant un enregistrement multi échelle du signal cérébral des patients. Ces électrodes sont constituées de macro-canaux permettant d'enregistrer l'activité de larges populations neuronales et de micro-canaux capables de capturer des signaux plus focaux, pouvant aller jusqu'à l'échelle du neurone unitaire. Nous avons construit un détecteur automatique de PEIs basé sur une nouvelle méthode de traitement du signal que nous avons baptisée Convolutional Kernel Density Estimation (CKDE). Nous avons également élaboré un détecteur automatique de FRs basé sur une approche écologique en trois étapes, imitant le travail du neurologue. Tous ces outils ont été incorporés à des interfaces graphiques utilisateurs (GUI) combinant les différentes fonctionnalités pour en permettre l'utilisation facile et efficiente. La détection des PEIs par CKDE offre la preuve de concept qu'une analyse orientée pixels de l'activité EEG peut être utilisée comme stratégie pour détecter des marqueurs intercritiques. Nous avons évalué cette méthode sur 10 minutes d'enregistrements chez un patient. Quinze PEIs ont été détectées automatiquement parmi lesquelles 13 vrais positifs et 2 faux positifs. Nos résultats principaux concernent toutefois la détection des FRs qui auraient à ce jour le plus grand potentiel dans le diagnostic des épilepsies pharmacorésistantes. Pour entraîner le CNN qui est une pièce maîtresse de notre détecteur, nous avons constitué une base de données de 4 954 FRs détectés manuellement chez 13 patients. Ce détecteur de FRs a été incorporé au logiciel que nous avons imaginé et créé, baptisé Ladybird, utilisé chez 29 patients pour détecter et traiter plusieurs milliers de FRs. Les avancées techniques et théoriques réalisées au cours de ce travail de thèse nous permettent d'envisager une utilisation à grande échelle de nos outils. Notre objectif est que les équipes médicales puissent en bénéficier directement, dans leur routine diagnostic. Un brevet a été déposé en vue d'un processus d'industrialisation
Improving Airline Pilots’ Visual Scanning and Manual Flight Performance through Training on Skilled Eye Gaze Strategies
No full textInternational audiencePoor cockpit monitoring has been identified as an important contributor to aviation accidents. Improving pilots’ monitoring strategies could therefore help to enhance flight safety. During two different sessions, we analyzed the flight performance and eye movements of professional airline pilots in a full-flight simulator. In a pre-training session, 20 pilots performed a manual approach scenario as pilot flying (PFs) and were classified into three groups according to their flight performance: unstabilized, standard, and most accurate. The unstabilized pilots either under- or over-focused various instruments. Their number of visual scanning patterns was lower than those of pilots who managed to stabilize their approach. The most accurate pilots showed a higher perceptual efficiency with shorter fixation times and more fixations on important primary flight instruments. Approximately 10 months later, fourteen pilots returned for a post-training session. They received a short training program and performed a similar manual approach as during the pre-training session. Seven of them, the experimental group, received individual feedback on their own performance and visual behavior (i.e., during the pre-training session) and a variety of data obtained from the most accurate pilots, including an eye-tracking video showing efficient visual scanning strategies from one of the most accurate pilots. The other seven, the control group, received general guidelines on cockpit monitoring. During the post-training session, the experimental group had better flight performance (compared to the control group), and its visual scanning strategies became more similar to those of the most accurate pilots. In summary, our results suggest that cockpit monitoring underlies manual flight performance and that it can be improved using a training program based mainly on exposure to eye movement examples from highly accurate pilots
Field report: deployment of a fleet of drones for cloud exploration
No full textInternational audienceDrones are commonly used for many civil applications and the procedures to operate them have evolved during the past years to make them accessible to those with limited piloting skills in several scenarios. However, the deployment of a fleet in the context of scientific research can lead to complex situations that require an important preparation in terms of logistics, permission to fly from authorities, and coordination during the flights. This paper is a field report of the flight campaign held end of January 2020 at the Barbados Island as part of the NEPHELAE project. The main objectives of the project were to fly into trade wind cumulus clouds to understand the microphysical processes involved in their evolution, as well as to provide a proof of concept of sensor-based adaptive navigation patterns to optimize the data collection. After presenting the overall flight strategy and the context of operation, the main challenges and the solutions to address them will be presented, to conclude with the evaluation of some technical evolution developed from these experiments
Widely linear FRESH receivers for cancellation of data-like rectilinear and quasi-rectilinear interference with frequency offsets
No full textInternational audienceWidely linear (WL) receivers have been developed in the past for single antenna interference cancel- lation (SAIC) of one rectilinear (R) or quasi-rectilinear (QR) data-like multi-user interference (MUI) or co-channel interference (CCI) in particular. The SAIC technology has been implemented in global system for mobile communications (GSM) handsets in particular and has been further analyzed for voice ser- vices over adaptive multi-user channels on one slot (VAMOS) standard. It remains of great interest for several current and future applications using R or QR signals, such as anti-collisions processing in radio frequency identification (RFID) or in satellite-AIS systems and to densify 5G and Beyond 5G (B5G) net- works through one dimensional signaling or over-loaded large MU-MIMO systems. It may be required to cancel the inter-symbol interference (ISI) of control and non-payload communications (CNPC) links of unmanned aerial vehicles (UAV) and the inter-carrier interference (ICI) of filter bank multi-carrier offset quadrature amplitude modulation (FBMC-OQAM), which are now candidate for B5G mobile networks. For these challenging applications, the development of enhanced WL filtering based SAIC or Multiple Antenna Interference Cancellation (MAIC) receivers for R and QR signals may be of great interest. Such a receiver, corresponding to a three-input WL frequency shift (FRESH) receiver, has been introduced recently for QR signals. However this WL receiver is not robust to a data-like MUI having a residual frequency offset (FO), which occurs for most of the previous applications. In this context, the paper first extends, for arbitrary propagation channels, the standards (for R and QR MUI) and the enhanced (for QR MUI) SAIC/MAIC WL receivers to MUI with a non-zero FO. Then, it shows the less efficiency of the two-input WL receiver for QR MUI with a non-zero FO and the performance improvement obtained with the three-input WL receiver. Finally, it analyzes, both analytically and by simulations, for R and QR MUI, the impact of the MUI FO on the performance of the proposed receivers. The results of the paper should allow the develop- ment of new powerful WL receivers for UAV CNPC links, anti-collisions AIS systems and for FBMC-OQAM networks in particular
The Impact of Automation on Air Traffic Controller’s Behaviors
No full textInternational audienceThis article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC B
From Lustre to Simulink: reverse compilation for verifying Embedded Systems Applications
No full textInternational audienceModel-based design is now unavoidable when building embedded systems and more specifically controllers. Among the available model languages, the synchronous dataflow paradigm, as implemented in languages such as Matlab Simulink or ANSYS Scade, has become predominant in the critical embedded system industries. Both of these frameworks are used to design the controller itself but also provide code generation means, enabling faster deployment to target and easier V\&V activities performed earlier in the design process, at model level.Synchronous models also ease the definition of formal specification through the use of synchronous observers, attaching requirements to the model in the very same language, mastered by engineers and tooled with simulation means or code generation.However, few works address the automatic synthesis of Matlab Simulink annotations from lower level models or code. We present here a compilation process from Lustre models to genuine Matlab Simulink, without the need to rely on external C functions or Matlab functions. This re-engineering is then used to validate a compilation tool-chain, mapping Simulink to Lustre and then C, thanks to equivalence testing and checking. This backward-compilation from Lustre to Simulink also provides the ability to produce automatically Simulink components modeling specification, proof arguments or test cases coverage criteria
Toolpath planning optimization for end milling of free-form surfaces using a clustering algorithm
No full textInternational audienceA two-step approach to address the toolpath planning optimization problem is introduced. The first step divides the surface into zones with similar local geometric properties in order to improve efficiency of toolpath planning algorithms. To do this, an unsupervised clustering algorithm (like K-means) is used on a mesh defined by isoparametric curves. The second step applies a toolpath planning algorithm to each zone, according to its optimal machining direction. This optimal direction is calculated using a black-box optimization software. The final goal is to enhance gradually the formulation of the optimization problem to obtain better and better results
A BENCHMARK OF THE GPS+GALILEO F9P RECEIVER
No full textInternational audienceStill very few publications relate test results of multiconstellation receivers since Galileo satellites massive launch by Ariane 5, leading to more than 20 operational space vehicles in orbit. One can refer to automotive receivers and smartphones test bench made during the European COST action SaPPART (Satellite Positioning Performance Assessment for Road Transport), but this was GPS only or GPS+Glonass. This article gives an overview of a test carried out in the city of Nantes, France, and its suburban area, with a brand new receiver, F9P, of the automotive range of Ublox. The dual constellation configuration GPS+Galileo is tested. This receiver is benchmarked with respect to the previous generation of Ublox LEA6T
Optimisation non linéaire en variables mixtes continues et discrètes pour des simulateurs de type de boîte noire
No full textIn recent years, there has been a considerable number of industrial applications that involve mixed variables and time-consuming simulators, e.g., at Safran Tech and IFPEN: optimal designs of aircraft engine turbine, of mooring lines of offshore wind turbines, of electric engine stators and rotors,... In these nonlinear optimization problems, derivatives of the objective function (and, possibly of the constraint functions) are not available and cannot be directly approximated. Another difficulty is that these problems involve heterogeneous nature variables: a varying number of components (integer variables), different materials (categorical variables, usually non-ordered), the presence or not of some components (binary variables), and continuous variables describing dimensions/characteristics of the structure pieces. This thesis aims to develop and adapt Derivative-Free Optimization (DFO) methods for different types of applications, including the optimal design of aircraft engines. In the first part, we focus on the development and adaptation of a DFO method to problems with continuous and mixed discrete variables exhibiting a cyclic-symmetry property. For that purpose, we introduce the necklace distance and tailor accordingly the trust-region constraints of the optimization problems. Before running our adapted method on a simplified simulation provided by Safran, we build a set of benchmark functions by transforming them into a set of cyclic-symmetry test functions. We run our method on these benchmark functions and on a Safran's simulated instance with a large number of repetitions to study the robustness of the method compared to other state-of-the-art methods. We also give a local convergence proof of our adapted method. In the second part, we focus on the design of experiments in mixed continuous and discrete variables space by extending the kernel-embedding distribution from continuous space to mixed discrete variables case. This part of the thesis is motivated by the need to improve the initialization phase of the optimization algorithm with a better exploration of the space of mixed variables, guided by the available prior information (types of variables, symmetry, correlations, ...). We illustrate the potential of the proposed approach in the more classical framework of meta-model function approximation for continuous and discrete mixed variables, and also for time series. Finally, we give ideas to improve the proposed optimization method for a better exploration of the space of design variables to avoid being trapped in local minima.Ces dernières années, un nombre considérable d'applications industrielles réelles impliquant des variables mixtes et des simulateurs coûteux en temps ont été réalisées, par exemple, chez Safran Tech et IFPEN, des conceptions optimales de turbine de moteur d'hélicoptère, de lignes d'amarrage d'éoliennes offshore, de stators et de rotors de moteurs électriques... Dans ces problèmes d'optimisation non linéaire, les dérivées de la fonction objectif (et, éventuellement des fonctions contraintes) ne sont pas disponibles et ne peuvent pas être directement approximées. Une autre difficulté est que ces problèmes impliquent des variables de nature hétérogène : un nombre variable de composants (variables entières), différents matériaux (variables catégorielles, généralement non ordonnées), la présence ou non de certains composants (variables binaires) et des variables continues décrivant les dimensions/caractéristiques des pièces de la structure. Cette thèse a pour but de développer et d'adapter des méthodes d'optimisation sans dérivées (ou DFO pour Derivative-Free Optimization) applicables à différents types d'applications, dont la conception optimale des moteurs d'hélicoptère. Dans la première partie, nous nous concentrons sur le développement et l'adaptation d'une méthode DFO aux problèmes avec variables mixtes continues et discrètes présentant une symétrie cyclique, caractéristiques présentes dans le problème d'optimisation des pales d'une turbomachine de moteur. À cette fin, nous introduisons une distance basée sur les colliers (necklace distance) et adaptons une distance basée sur les colliers pour une méthode d'optimisation du type région de confiance. Avant d'appliquer notre méthode à un cas applicatif simplifié fourni par Safran, nous construisons un ensemble de fonctions tests issues de la littérature que nous adaptons pour obtenir un ensemble de problèmes mixtes à symétrie cyclique. Notre méthode est évaluée sur ces cas tests et comparée à des méthodes d'optimisation sans dérivées de l'état de l'art. Nous donnons également une preuve de convergence locale de notre méthode adaptée. Dans la deuxième partie, nous nous consacrons à la planification d'expériences dans un espace mixte (variables continues et discrètes) en étendant à cet espace mixte les approches basées sur des méthodes à noyaux pour estimer des distributions de probabilité. Cette partie de la thèse est motivée par le besoin d'améliorer la phase d'initialisation de l'algorithme d'optimisation. Le but étant de permettre une meilleure exploration de l'espace des variables mixtes, guidée par les informations a priori disponibles (types de variables, symétrie, corrélations,...). Nous illustrons également le potentiel de l'approche proposée dans le cadre plus classique de l'approximation d'une fonction par un méta-modèle pour des variables mixtes continues et discrètes mais aussi pour des séries temporelles. Enfin, nous donnons des pistes d'amélioration de la méthode d'optimisation proposée pour une meilleure exploration de l'espace des variables de conception pour éviter d'être piégé dans des minima locaux