Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Engineering science >> Mathematics - Numerical analysis - Simulation
11 proposition(s).

Analog to Digital Converter for Neural Network based Acoustic Detection System for IoT Systems

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Architectures Intégrées Radiofréquences

01-09-2018

SL-DRT-18-0494

dominique.morche@cea.fr

The purpose of this PhD is to develop a new analog to digital converter whose output will be some trains of spike. This waveform is particularly adapted to the neural network processing. A joint optimization between the ADC and the neural network will be done by the PhD. Signal Detection and Classification is becoming a key function in the internet of things to extract useful information from the environment. For such functionality, neural network based processing is becoming more and more interesting. However, the power efficiency is often limited by the analog to digital interface which is mandatory. Therefore, more power efficient analog to digital interface are required and their power consumption should adapt itself to the application requirements. That is the reason why, the purpose of this PhD is to develop a new analog to digital converter whose output will be some trains of spike. This waveform is particularly adapted to the neural network processing. A joint optimization between the ADC and the neural network will be done by the PhD, in collaboration with a Post-Doc who will be working on the digital part. The design will exploit the 28nm FD-SOI technology developed by STMicroelectronics. Several circuit will be design, fabricated and tested. The objective at the end is to build a demonstrator able to distinguish audio signals. Industrials use cases will be considered. The Phd will be done in collaboration with IMT Atlantique. The PhD will be asked to present his work in the scope of collaborative European project.

Télécharger l'offre (.zip)

Air pollution control by means of a water layer

Département Microtechnologies pour la Biologie et la Santé (LETI)

Laboratoire Biologie et Architecture Microfluidiques

01-10-2018

SL-DRT-18-0616

jean-maxime.roux@cea.fr

Air pollution, especially urban air pollution, is a public health problem leading in France to nearly 50 000 deaths per year. The PhD subject deals with the design of a new urban clean-up system based on wet electrostatic precipitation. Air purifiers based on this principle are usually intended for an industrial use. The PhD will focus on a multiphysical numerical simulation of such a device, but adapted to an urban deployment, starting with the central and difficult problem posed by the stability of an air/water interface in an intense electric field. While being based on this simulation, the final challenge is to develop a numerical optimization of the system aiming at a significant reduction of its size and an appropriate integration of the toxic gas / airborne particles sensors developed at CEA GRENOBLE/Leti/DTBS. Experimental studies carried out at CEA will be guided by the obtained numerical results which will in return be validated.

Télécharger l'offre (.zip)

Applying machine learning to improve Intrusion Detection Systems

DPACA (CTReg)

Autre

01-09-2018

SL-DRT-18-0617

pierre-alain.moellic@cea.fr

The proliferation and growing complexity of cyber-attacks targeting the networks of companies, institutions or industrial infrastructures is a major security issue. Today, it is essential to propose technological solutions to detect complex and usually new attacks more particularly for critical infrastructures such as Cyber Physical Systems (CPS) gathering strong operational constraints. Among the available security tools, intrusion detection systems (IDS) rapidly become indispensable solutions such as traditional firewalls or antivirus. However, the available solutions cannot completely thwart current threats mainly because of a detection paradigm that is focused on known attacks (misuse-based or signature-based IDS). The future of these systems go through the development of other approaches (anomaly-based IDS) and the use of analysis and modelling tools based on Machine Learning. A lot of academic works have been proposed in this sense, supported by the strong emulation in ?artificial intelligence?. However, proposed technologies suffer from a lack of real data enabling an efficient evaluation of the performances. In the highly critical context of CPS which we need to precisely define the architectures, the supervision processes and threat models, the PhD aims at developing innovative IDS solutions (using well-known open source platforms) using approaches based on Machine Learning and using real data (from CEA Cadarache). The proposed solutions will have to meet strong performance requirements (accuracy rate, false-positive/false-negative rates) to demonstrate the pertinence of these approaches for real infrastructures.

Télécharger l'offre (.zip)

Mesoscopic scale simulation of the Barkhausen effect for the characterization of steels

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Simulation et Modélisation en Electro-magnétisme

01-09-2018

SL-DRT-18-0642

anastasios.skarlatos@cea.fr

Barkhausen noise is more and more used as a measure of the health state of magnetic materials. It is indeed strongly correlated to materials microstructure, stress level and chemical composition for instance. In spite of its great practical interest, this measure is often hard to interpret due to the large number of underlying physical phenomena. The development of efficient and accurate modelling tools is thus necessary to enhance the understanding of measurement and access to more quantitative estimations of characteristic quantities, such as the level of stress or a rate of chemical component. From the modelling point of view, the problem to solve is complex due to its multi-scale nature. Existing approaches can be divided in two families: those based on Monte Carlo methods to get a very fine description at the spin level, and those labelled as mesoscopic ones, aiming at solving a magnetostatic problem at the scale of the magnetic domains. In these latter approaches, Maxwell equations are solved considering a simplified configuration of domains in terms of geometry and displacements of domain walls. This PhD subject consists in implementing an optimized simulation tool for the characterization of steels, based on a mesoscopic approach. This tool will exploit empirical considerations on the distribution and dynamic behavior of domain walls in view of deriving macroscopic signals measured in practice and studying the statistics of characteristic parameters involved. Magnetostatic simulations will be carried out with a 3D numerical solver based on Finite Integration Technique (FIT) developed at CEA LIST. The representativeness of the unitary calculation will be the key to the validity of the statistical procedure leading to macroscopic signals. Theoretical results will be compared to experimental data obtained in laboratory controlled conditions by partners of laboratoire Roberval (Université de Technologie de Compiègne, UTC), involved in the PhD work.

Télécharger l'offre (.zip)

Accurate and robust estimation of the PEMFC ageing state Bayesian observers using a model-based approach

Département de l'Electricité et de l'Hydrogène pour les Transports (LITEN)

Laboratoire Electronique avancée, Energie et Puissance

01-10-2018

SL-DRT-18-0665

vincent.heiries@cea.fr

PHM (Prognostics and Health Management) represents a real opportunity to improve fuel cell performance and extend the life of fuel cells. This field of study has recently gained much interest. The main goal is to make optimum use of the data measured by all available sensors in order to evaluate the specific indicators of PEMFC ageing and possibly modify the operation of the fuel cell in order to optimize its lifetime. The proposed PhD is part of a model-based approach and will be based on the expertise in fuel cell modelling developed at the Modelling Laboratory. An on-line estimator of the ageing state of the fuel cell will be developed. The proposed observer presents the characteristic of combining a state model derived from the MEPHYSTO fuel cell model with the different data sensors available (voltage, current, pressure, temperature). The envisaged method makes it possible to jointly estimate the state variables, and in particular the ageing state, as well as to update the model parameters. Given the nature of the state variables to be estimated, we will move towards sophisticated observers adapted to non-linear and non-Gaussian problems in order to obtain a solution approaching the optimal Bayesian estimate.

Télécharger l'offre (.zip)

Optimisation of energy management strategy for fuel cell hybrid véhicles by combinatorial optimization methods using multi-physics models of performance and degradation

Département de l'Electricité et de l'Hydrogène pour les Transports (LITEN)

Laboratoire Modélisation multi-échelle et suivi Performance

01-10-2018

SL-DRT-18-0856

ramon.naiff-dafonseca@cea.fr

The hybrid architecture requires the management of the components in order to improve some characteristics when comparing to conventional solutions. The power split strategy among the components allows to increase the performances of the system by the minimization of the energy consumption and/or increasing the durability of the system, while considering the mission requirements and technological constraints. Several methodologies has been used to develop an approach to obtain the energy management strategy with important results, but also with some limits related to the application. This PhD thesis proposition aims to deal with the indicated problems by applying a methodology able to solve a problem with different sources (battery, hydrogen fuel cells, super capacitors, etc.), different criteria (energy consumption, durability, cost, etc.) and several state dimensions (time, SOC, voltage, temperature, etc.) in an optimal way while considering a time calculation adapted to the application (sizing offline calculation, embedded implementation, etc.). The combinatory optimization methods are considered, at first sight, as valid options thanks to their capacity to solve complex and nonlinear problems that are used present in multi-sources and multi-dimensions hybrid applications. In the State of the Art, this type of methodology have shown the possibility to perform global optimization with less calculation effort and less memory allocation than the Dynamic Programming Method. Moreover, this methodology is less depend on the discretization of the problem as Dynamic Programming is. The main objective of this thesis is to develop an optimization methodology that combines the models of the system (performance models, controls and degradation) to the combinatory optimization strategy in a formal optimization problem.

Télécharger l'offre (.zip)

Compact modeling of MOSFETs operating at cryogenic temperatures

Département Composants Silicium (LETI)

Laboratoire de Simulation et Modélisation

01-10-2018

SL-DRT-18-0883

thierry.poiroux@cea.fr

Recent experimental demonstrations of silicon-based Qbits pave the way to the fabrication of quantum computing circuits with standard nanoelectronics technologies. However, the design of such Qbit-based functional circuits requires a relevant design environment. In particular, designers need compact models able to reproduce the behavior of MOSFETs operating at cryogenic températures. CEA-Leti has developed a physics-based compact model dedicated to FDSOI technology, called Leti-UTSOI, that is used in industrial design kits. This model has been conceived to fulfill the requirements of applications operating close to ambient temperature (typically -45° to +125°). Therefore, new developments are required in order to extend the predictability of Leti-UTSOI down to ultra-low températures. These developments are essential in order to simulate, optimize and validate the design of circuits at cryogenic temperature. Moreover, they will be implemented in other compact models dedicated to bulk MOSFET technologies or to finFET and nanowire device architectures. The availability of these models, able to reproduce the behavior of various transistor architectures at very low températures, will allow very useful benchmarks of MOSFET technologies for cryogenic applications.

Télécharger l'offre (.zip)

Performance demonstration of Guided Wave based Structural Health Monitoring for aerospace

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Méthodes CND

01-10-2018

SL-DRT-18-0918

olivier.mesnil@cea.fr

Access to space is a critical stake of the first half of this century, demanding the conception of new low-cost and reusable space launchers leading to a new kind of demand for structure monitoring and maintenance. Structural Health Monitoring (SHM) is a group of methodologies aiming at monitoring large and potentially complex structures by embedding sensors permanently on or within it. SHM is particularly promising for aeronautical and aerospace structures to provide early defect detection and optimize the use. Guided Wave (GW) based SHM relies on the use of guided elastic waves to interrogate the structure and detect flaws as change in the GW propagation. Even though multiple in-lab prototypes of GW-SHM system do exist, the main barrier to a mainstream adoption of the technology is the difficulty to demonstrate the performances of such a system. Indeed, the performances must be quantified and certified before any usage of such technologies in the demanding aerospace and aeronautical industries. The objectives of this thesis is to develop, based on simulation tools for GW simulation, a methodology to quantify the performances of a GW-SHM system. This goal will require the following tasks: After getting used to the simulation tools and the SHM imaging methodologies available in the laboratory, a statistical approach to quantify the performances of an SHM system based on the simulation will be developed. Experimental trials will then be conducted to validate the approach and eventually improve the gap between the simulation and the experiment. This thesis is part of the collaboration CEFIPRA program between France and India (CEFIPRA: Centre franco-indien pour la Promotion de la Recherche avancée).

Télécharger l'offre (.zip)

3D Objects discovery in 3D scene

DPLOIRE (CTReg)

Autre

01-10-2018

SL-DRT-18-1039

anthony.mouraud@cea.fr

Object detection and localization in images is a problem studied since many years. The latest technological developments now allow the real-time acquisition of depth data coupled to color data (RGBD). At the same time, modern machine computing capabilities and intelligent image processing methods have led to significant advances in the detection / localization of 2D objects with many different approaches (bounding boxes, contours, from CAD models ...). An important step is being taken in recent years with the research conducted to directly extract the volume of detected objects and their position in 3D. These works are still in their infancy, but the first results are encouraging, both from 2D images (eg DeepManta) and from 3D images (eg Deep Sliding Shapes). However, there remain several identifiable scientific / technological barriers before allowing the democratization of this type of approach for the automatic extraction of objects in potentially unknown scenes. The objective of this work is to identify the current approaches of detection / localization of 3D objects, to target their weaknesses and work on new processing technologies to mitigate them. Moreover, the object discovery in unknown environments and the inference of the operator's intention by observation / location of his attention are two areas of interest that this work aims at addressing. Beyond their applications for demonstration learning, the software bricks resulting from this project can also be reused for other applications such as augmented reality ("smart" scanning, etc.), surveillance or mobile mobility for example.

Télécharger l'offre (.zip)

Inverse reinforcement learning of a task performed by a human

DPLOIRE (CTReg)

Autre

SL-DRT-18-1047

laurent.dolle@cea.fr

Learning from demonstration involves an agent (e.g., a robot) learning a task by watching another agent (e.g., a human) performing the same task. It often uses reinforcement-learning methods to improve the robot's ability to perform a task in new situations (i.e., generalization). These methods involve providing a positive reinforcement (i.e., a reward) when the outputs of the algorithms help achieving the task, but require a human designed reward function. The more the task is complex the more difficult is the reward function to design, but it can be learned from a series of examples with methods called inverse reinforcement learning. The use, jointly or not, of these techniques has shown encouraging results, but which are limited to toy examples and cannot be adapted as such to tasks more representative of the industrial environment. During the thesis, the PhD student will analyze and test state-of-the-art previous works. S/He will then propose a method, combining inverse reinforcement learning to other algorithms (e.g., generative adversarial networks, GAN), so that the robot will understand the task performed by the operator (with as little explanation from the operator as possible), and will generalize enough to make the robot robust to dynamic environments (obstacles, moving objects?). This method should be suited for a "pick and place" task in an industrial environment and ensure a reasonable enough learning period (information a priori, feedback from the operator) for tasks of medium complexity.

Télécharger l'offre (.zip)

Machine Learning for a precision agriculture

DLORR

01-11-2018

SL-DRT-18-1060

ulysse.marboeuf@cea.fr

This PhD is at the interface between agriculture and Machine Learning. The project is based on a collaboration between the CLAAS company based in Woippy (Moselle) specialized in the manufacture of "high-end" agricultural Equipment and CEA Tech in Metz. This PhD is part of the statistical modeling of an agricultural press system. It aims to design a parametric statistical model by supervised learning, to automate the compression procedure of the biological material and help the farmer in this task. This model must meet physical and environmental constraints.

Télécharger l'offre (.zip)

Voir toutes nos offres