Post-doc: CNN neural network – managing data uncertainty in the learning database.
The aim is to develop algorithms able to take into account the uncertainty in the learning database of neural networks. The project fits into the context of the dynamic state estimation of liquid-liquid extraction and benefits of its knowledge-based simulator as well as industrial data. Indeed, the status of an industrial chemical process is accessible through operating parameters and available monitoring measures. However, the measures being inherently associated with uncertainty, it is necessary to make the data consistent with process knowledge. Therefore, the goal is to find the best data set of operational parameters (input of the knowledge-based simulator) to provide the model to estimate the real process state known through monitoring measures (output of the knowledge-based simulator). A convolutional neural network (CNN) is being developed in another postdoctoral project to solve the inverse problem to find the best input thanks to the measured output. A consistent set of operating parameters is going to be obtained and state of the process is going to be known during the dynamic regime of the liquid-liquid extraction process. This first step is to evaluate the impact of the uncertainty of operational parameters on the outputs of the knowledge-based model. This step will need to connect the knowledge-based model to URANIE, internal platform developed by CEA ISAS. This knowledge must be taken into account in the second part of the project. The uncertainty observed on the outputs should be taken into account in the learning loop to improve the estimation of the operational parameters by the CNN. The impact of these uncertainties on the CNN computed results must be assesed in order to trust the ability of the CNN to estimate the state of the process.
Through this project, we are at the heart of the thematic of digital simulation for the best control of complex systems.
Deployment of distributed consensus protocols on blockchains with Smart Contract
The aim is to implement various distributed consensus protocols on both public and private blockchain platforms supporting Smart Contracts technology. The techniques based on Proof-of-Stake and token management will be analyzed and their level of security will be evaluated in terms of energy consumption and quality of the distribution of the trust in the system. The techniques to verify the transactions of the blockchain Ethereum will be implemented, as well as other algorithms, lighter and that consume less energy, dedicated to "private" blockchains where users are authenticated. The platform Hyperledger will be used to test the various distributed consensus protocols. New algorithms will be proposed and the solutions will be deployed for applications in the field of the Internet of Things.
Machine learning technics and knowledge-based simulator combined for dynamic process state estimation
This project aims to estimate the real state of a dynamic process for liquid-liquid extraction through the real data record. Data of this kind are uncertain due to exogenous variables. They are not included inside the simulator PAREX+ dedicated to the dynamic process. So, the first part of the project is to collect data from simulator. By this way the operational domain should be well covered and the dynamic response recorded. Then, the project focuses to solve the inverse problem by using convolutionnal neural networks on times series. Maybe a data enrichment could be necessary to perfect zones and improve estimations. Finally, the CNN will be tested on real data and integrate the uncertainty inside its estimations.
At the end, the model built needs to be used in operational conditions to help diagnosis and improve the real-time control to ensure that the dynamic observed is the one needed.
Correlative X-ray and ToF-SIMS tomography Data fusion of 3-D data sets from X-ray and ToF-SIMS tomography
The nanocaracterisation platform of the CEA Grenoble has recently installed 2 state-of-the-art tools for 3-D imaging with 100 nm resolution: X-ray tomography in a SEM and time of flight secondary ion mass spectrometry (ToF-SIMS) assisted by focused ion milling (FIB). X-ray tomography delivers non-invasive 3-D images of the internal morphology of an object whilst ToF-SIMS is able to map the local composition in 3-D. We aim to combine the two techniques to perform quantitative 3-D investigations of objects such as copper pillars for microelectronics or silicon electrodes for Li battery applications.
The proposed research subject is data analysis orientated. Some simulation work may be performed to implement and test existing 3-D data fusion methods with a view to adapting and improving them. The candidate will assist with the experimental measurements and be responsible for treating the data with the chosen protocols. The candidate should be pragmatic, at ease with applied mathematics and have good programming skills. These will be essential in understanding and manipulating the fusion and reconstruction algorithms, from the simplest, to the increasingly advanced (prior information, superiorisation, Bayesian fusion)
The candidate will have completed a PhD in physics and have good computer (Python, Matlab, C) and image treatment skills, or a PhD in mathematics/computational science with an interest in applications. The candiate will need to interface with a multidisciplinary team, and be receptive to new ideas. The candidate will be proficient in both written and spoken English in order to communicate with the team and to disseminate their results in articles or at conferences.
Error Coding Driven Synthesis of Combinational Circuits from Unreliable Components
With the advent of nanoelectronics, the reliability of the forthcoming circuits and computation devices is becoming questionable. Indeed, due to huge increases in density integration, lower supply voltages, and variations in the technological process, MOS and emerging nanoelectronic devices will be inherently unreliable. As a consequence, the nanoscale integration of chips built out of unreliable components has emerged as one of the most critical challenges for the next-generation electronic circuit design. To make such nanoscale integration economically viable, new solutions for efficient and fault-tolerant data processing and storage must now be invented.
This post-doctoral position aims at investigating innovative fault-tolerant solutions, at both device- and system-level, that are fundamentally rooted in mathematical models, algorithms, and techniques of information and coding theory. Investigated solutions will build on specific error correcting codes, able to provide reliable error protection even if they themselves operate on unreliable hardware. The goal is to develop the scientific foundation and provide a first proof-of-concept, as an essential condition for bringing about a paradigm shift in the design of future nanoscale circuits.
Developement of a simulation platform for the energy systems
The evolution of power systems towards smart-grids, including a high share of renewable generation which can be combined with storage systems, lead to an increased complexity for designing and optimizing these systems. This leads to a need for new modeling and simulation tools, which have to manage different energy sources, different energy vectors and different technologies for energy conversion. Moreover, such simulation tools will be used to optimize the system sizing and to design energy management strategies.
The objective of this project is to design the software architecture for the simulation platform, which will be in ad equation to the previously mentioned needs. Such software will be organized in order to maximize the transfer towards industrial partners. The software will be able to support multi-energy systems, and will leave the possibility for the user to implement its own component models or energy management strategies.
The project is focused on the simulation platform architecture, and on the architecture model. This architecture will be used as a base for the development of a software. The objective of the given project is not to cover all the applications but rather to validate the architecture through a given application.
Robust path-following solvers for the simulation of reinforced concrete structures
Path-following procedures are generally employed for describing unstable structural responses characterized by ``snap-backs'' and/or ``snap-troughs''. In these formulations, the evolution of the external actions (forces/displacements) is updated throughout the deformation process to fulfill a given criterion. Adapting the external loading during the calculation to control the evolution of the material non-linearities is helpful to obtain a solution and/or to reduce the number of iterations to convergence. This second aspect is of paramount importance, especially for large calculations (at the structural scale). Different path-following formulations were proposed in the literature. Unfortunately, an objective criterion for choosing one formulation over another for the simulation of reinforced concrete (RC) structures (in the presence of different and complex dissipation mechanisms) still needs to be made available. The proposed work will focus on the formulation of path-following algorithms adapted to simulate RC structures.
ACCELERATING a DSN SWEEP KERNEL ALGORITHM FOR NEUTRONICS BY PORTING ON GPU.
In the framework of the Programmes Transversaux de Compétences (PTC or literally Cross-XXX Programme), the DES/ISAS/DM2S/SERMA/LLPR and the CEA-DIF are both working on the porting of deterministic neutron transport codes on GPU.
The DM2S within the Energies Direction (DES) is responsible for research and development activities on the numerical methods and codes for reactor physics, amongst which the APOLLO3® code. The neutronics laboratory of CEA-DIF is responsible for developing tools for deterministic methods in neutronics for the Simulation programme.
These two laboratories are actively preparing for the advent of new generation of supercomputers where GPU (Graphical Processing Units) will be predominant. Indeed, the underlying numerical problems to be solved along with the working methodology as well as the conclusions and experience which will be obtained from such studies may be rationalised between both laboratories. Thus, this work has given rise to this postdoctoral position which will be common to both teams. The postdoctoral researcher will be formally based at SERMA at CEA Saclay, with nevertheless regular meetings with the CEA-DIF scientists.
The postdoctoral research work is to study the acceleration of a toy model of a 3D discrete ordinates diamond-differencing sweep kernel (DSN) by porting the code on GPU. This work hinges on porting experiments which have previously been carried by both teams following two different approaches: a ‘’high-level’’ one based on the Kokkos framework for DES and a ‘’low-level’’ approach based on Cuda for CEA-DIF.
Neutronic thermal-hydraulic coupling in heterogeneous Sodium Cooled Fast Reactor (SCFR)
Within the frame of ASTRID (Sodium cooled Fast Reactor) prototype development, update of calculation methodologies using new generation of codes benefiting from High Performance Computing (HPC) and advanced coupling capabilities is underway. These methods are expected to be integrated in ASTRID safety demonstration. In particular, development of coupled neutronics/thermal-hydraulics/fuel mechanics methodologies during accidental transients is underway.
Coupling Neutronics and thermal-hydraulics in double phase flow conditions (either sodium + vapor sodium or sodium + other gaz) can be used for:
• Loss of Flow transients (LOF, sodium + vapor sodium)
• Gas insertion transients.
This coupling is of special interest with cores strongly relying on axial leakage for safety consideration (like CFV cores [ICAPP11]).
The work proposed is to further develop the implementation of 3D coupling with state of the art CEA codes (APOLLO3, FLICA, CATHARE, TRIO etc.) to analyze the two type of transients stated above.
Development of Monte-Carlo methods for the simulation of radiative transfer: application to severe accidents
This post-doctoral subject concerns the development of Monte-Carlo ray-tracing methods for modeling radiation heat transfer in the context of severe accidents. Starting from a well-developed software framework for Monte Carlo simulation of particle transport in the context of reactor physics and radiation protection, we will seek to adapt existing methods to the problem of radiative heat transfer, in a high-performance computing framework. To do this, we will develop a hierarchy of approximations associated with radiative heat transfer that are intended to allow the validation of simplified models implemented in the context of the numerical simulation of severe accidents in nuclear reactors. Focusing on algorithm and simulation performance, this work is intended to be a "proof of principle" of the possible software mutualization around the Monte-Carlo method for particle transport on the one hand and radiative heat transfer on the other hand.