Automatic machine learning identification of nanoscale features in transmission electron microscopy images
Imaging nanoscale features using transmission electron microscopy (TEM) is key to predicting and assessing the mechanical behaviour of structural materials in nuclear reactors or in the fields of nanotechnology. These features, visible by phase contrast (nanobubbles) or diffraction contrast (dislocation loops or coherent precipitates), are prime candidates for automation. Analysing these micrographs manually is often tedious, time-consuming, non-universal and somehow subjective.
In this project, the objective is to develop a Python-based framework for data treatment of transmission electron microscopy (TEM) images.
Machine Learning approaches will be implemented in order to tackle the following tasks:
- Data collection: The success of any machine learning approach is linked to the database quality. In this project, a huge database is available. Four microscopists are involved in the project and will continuously enrich the database with images containing easily recognizable features.
- Denoising and finding the defect contour both through existing open-access software and in-house developed descriptors. Representative ROI (region-of-Interest) will be generated on images.
- Design of the Convolutional Neural Network (CNN) Architecture and model training: A collective feature map will be generated for the entire images in order to identify some representatives ROI. Each ROI is then overlaid to the original feature map and is passed to the CNN for individual region classifications. Secondly, recent advances in image segmentation will be placed in the core engine of the workflow.
- Model performance metrics: The aim is to reach a compromise between the training time and the detector performance.
The process will be applied to nanometer-sized features formed under irradiation in nuclear oriented materials (Co-free high entropy alloys (HEA), UO2) and precipitates in materials with a technological interest (coherent Cr precipitates in Cu).
Hydrolysis energy distribtution in model glasses using molecular simulation and Machine Learning
The objective of the project is to develop a tool based on molecular simulations combined with Machine Learning to estimate rapidly the distributions of hydrolysis and reformation energies of the chemical bonds on the surface of alumino silicate glasses(SiO2+Al2O3+CaO+Na2O).
The first step will consist in validating the classical force fields used to prepare the hydrated SiO2-Al2O3-Na2O-CaO systems by comparison with ab initio calculations. In particular, metadynamics will be used to compare classical and ab initio elementary hydrolysis mechanisms.
The next step will consist in performing « Potential Mean Force » calculations using the classical force fields to estimate distributions of hydrolysis and reformation energies on large statistics in few glass compositions. Then by using Machine Learning and atomic structural descriptors, we will try to correlate local structural characteristics of the chemical bonds to the hydrolysis and reformation energies. Methods such as Kernel Ridge Regression, Random Forest or Dense Neural Network will be compared.
At the end, a generic tool will be available to rapidly estimate distributions of hydrolysis and reformation energies for a given glass composition.
Development of irradiation resistant silicon materials and integration in photovoltaïcs cells for space applications
Historically, photovoltaic (PV) energy was developed together with the rise of space exploration. In the 90’s, multijunction solar cells based on III-V materials progressively replaced silicon (Si) cells, taking advantage of higher efficiency levels and electrons/protons irradiation resistance. Nowadays, the space environment is again looking at Si based PV applications: request of higher PV power, moderated space mission lengths, cost reduction issues (€/W Si ~ III-V/500), higher efficiencies p-type Si PV cells… Solar cells are exposed to cosmic irradiation in space, especially to electrons and protons fluxes. The latter’s affect the cells performances, essentially because of bulk defect formations and charge carrier recombination. In order to use Si based solar cells in space, we need to increase their irradiation resistance, which is the main goal of this post-doc position. To do so, the work will first consist in elaborating new Si materials, with increased irradiation resistance. Compositional aspects of the Si will be modified, particularly by introducing elements limiting the formation of bulk defects under irradiations, developing electrical passivation properties. The electronic properties of the materials will be deeply characterized before and after controlled irradiation. Then, this Si material will be used to fabricate heterojunction solar cells. Their performances will be evaluated again before and after irradiation. Such experimental work could be supported by numerical simulation at the device scale.
HPC simulations for PEM fuel cells
The goal is to improve TRUST-FC software -a joint development between LITEN and DES institutes at CEA- for detailed full 3D simulation of hydrogene PEM fuel cells and to run simulations on whole real bipolar plate geometries. Funded by AIDAS virtual lab (CEA/Forshungs Zentrum Juelich), a fully coupled electro-chemical, fluidic and thermal model has been built, based on CEA software TRUST. The model has been benchmarked against its FZJ counterpart (Open fuelcell, based on OpenFoam). The candidate will adapt the software and toolchain to larger and larger meshes up to billion cells meshes required to model a full bipolar plate. Besides, he will introduce two phase flow models in order to address the current technological challenges (local flooding or dryout). This ambitious project is actively supported by close collaboration with CEA/DES and FZJ.
Modeling of faults on low voltage DC networks in buildings, towards fault detection algorithms
The development of the use of renewable energies and energy storage as well as the progress made by power electronic components are gradually leading to a rethinking of the architectures of low voltage electrical distribution networks in buildings. These developments will allow the development of direct current or mixed alternating-direct current networks supplied by static converters. On this type of network, faults become more difficult to manage due to the power sources used. Indeed, the usual signatures of the short-circuit or the overload are no longer the same and will vary according to the converters used and the architecture of the network. For this, it is necessary to identify, by simulation, the most suitable protection topologies (by neutral systems for example) and to identify the typical fault signatures. Ultimately, these signatures will provide optimum detection devices.
Lean-Rare Earth Magnetic materials
The energy transition will lead to a very strong growth in the demand for rare earths (RE) over the next decade, especially for the elements (Nd, Pr) and (Dy, Tb). These RE, classified as critical materials, are used almost exclusively to produce NdFeB permanent magnets, and constitute 30% of their mass.
Several recent international studies, aiming to identify new alloys with low RE content and comparable performances to the dense magnetic phase Nd2Fe14B, put hard magnetic compounds of RE-Fe12 type as advantageous substitution solutions, allowing to reduce more that 35% of the amount of RE, while keeping the intrinsic magnetic properties close to those of the Nd2Fe14B composition.
The industrial developments of the RE-Fe12 alloys cannot yet be considered due to the important technological and scientific challenge that remain to be lifted in order to be able to produce dense magnets with resistance to demagnetization sufficient for current applications (coercivity Hc > 800 kA/m).
The aim of the post-doctoral work is to develop Nd-Fe12 based alloys with optimized intrinsic magnetic properties and to master the sintering of the powders in order to obtain dense magnets with coercivity beyond 800 kA/m, to fulfil the requirements of the applications in electric mobility. Two technological and scientific challenges are identified:
- understanding of the role of secondary phases on the coercivity. This will open the way to the implementation of techniques called "grain boundary engineering", well known for the NdFeB magnets to have remarkably improved the resistance to demagnetization.
- mastering the sintering step of these powders at low temperature (< 600°C) in order to avoid the decomposotion of the magnetic phase by grain boundary engineering
High efficiency silicon cell irradiations for space
Historically, photovoltaics was developed in conjunction with the growth of space exploration. During the 90's, III-V multi-junction solar cells were progressively replaced silicon, for their superior performance & radiation hardness. Today, the context is favorable to a revival of space Si: increasing PV power needs, missions with moderate durations & constraints (LEO), very low cost & high performance terrestrial Si cells (p-type > 26% AM1.5g). However, for Si cells, conventional irradiation ageing methods & sequences (ECSS) are less appropriate. As the literature mainly comes from 80s - 90s, it is necessary to revisit the topic for the latest generation of passivated contacts Si cells (developed at CEA INES) and the unique double beam irradiation facilities of JANNuS platform - CEA Saclay.
This work is part of the SiNRJs project, at the interface between two CEA departments, dealing with space photovoltaics & materials irradiation. The scientific & technological approach adopted: 1. fabrication of passivated contact Si cells (HeT and/or Poly-Si) 2. Si cells optoelectronic characterizations before irradiation (IV AM1.5/AM0, EQE, etc.) 3. Cells & samples proton irradiations, in situ characterizations (Raman & El) 4. Ex situ characterizations after irradiations (IV AM1.5/AM0, EQE, etc) 5. Results analysis and synthesis. From a scientific point of view, the key issues to be addressed concern the understanding of the mechanisms/dynamics of defect creation/healing under this double electronic and ballistic excitation.
Rhelogical properties of molten crystallized glass
Formulation of nuclear waste conditioning glass results from a compromise between waste loading, glass technological feasibility and its long-term behavior. Up to now borosilicate glasses formulated at CEA and elaborated at La Hague plant by Orano to condition nuclear waste are homogeneous when molten. That means that today glass formulation is determined such as solubility limits of each constituting elements of waste aren’t exceeded in order to avoid phase separation (implying typically Mo, S, P) and/or crystallization (implying typically Fe, Ni, Cr, Zn, Al, Ce, Cs, Ti…) leading to a two-phase molten glass (liquid-liquid or liquid-solid).Today CEA would like to explore the impact of solid particles in suspension in the molten glass and in the final glass canister on respectively the glass technological feasibility and its long-term behaviour.
The proposed study focuses on the molten glass technological feasibility. The presence of solid heterogeneities in the melt is known to lead to a modification of some of its physical properties – notably its rheology, as well as thermal and electrical conductivities, and can generate settling phenomena. Yet these properties are in the heart of vitrification process control and modelling. This study will then investigate the impact of crystals in the molten glass on vitrification process control and modelling.
Synthesis by 3D printing of functionnalized geopolymer membrane for the treatment of complex radioactive effluents.
In the field of the treatment of liquid radioactive wastes on solid supports, the development of new composite materials synthetized by 3D printing under filtre shape is of primary of importance to decontaminate some radioactive effluents.
In this phD proposal, we propose to develop a membrane allowing to produce, from effluent containing somes traces of micronic solids in suspension and ionic species, a clarified effluent compatible with a nuclear outlet pipe. The challenge is to study the shaping of a material in a form of a filtration membrane allowing to trap in a single step an effluent containing some solids in suspension and some ionic species. In order to develop both functionnalities, 3D printing will be used to synthetise multiscale porous ceramic composites such as some geopolymers functionnalized by a selective adsorbants. The candidate, mainly based at CEA/ISEC Marcoule, could first formulate a functionnalized geopolymer paste with suitable rheological properties compatible with the constraints of the 3D printing process. A cross-flow filtration membrane with a controled macroporous network will be then printed by optimizing the geometry of the mesh. Finally, some sorption and cross-flow filtration tests will be performed on some model effluents containing calibrated solid in suspension and ions of interest such as Cs and Sr. The relevance of the printed membrane architecture will be assessed in relation to the capture of the solids and radioelements.
The candidate must have skills in the field of rheology, process and modeling. From this research work, job opportunities either in the field og 3D printing of materials or in the field of liquid waste treatment and depolution are potential options.
Robotics Moonshot : digital twin of a laser cutting process and implementation with a self-learning robot
One of the main challenges in the deployment of robotics in industry is to offer smart robots, capable of understanding the context in which they operate and easily programmable without advanced skills in robotics and computer science. In order to enable a non-expert operator to define tasks subsequently carried out by a robot, the CEA is developing various tools: intuitive programming interface, learning by demonstration, skill-based programming, interface with interactive simulation, etc.
Winner of the "moonshot" call for projects from the CEA's Digital Missions, the "Self-learning robot" project proposes to bring very significant breakthroughs for the robotics of the future in connection with simulation. A demonstrator integrating these technological bricks is expected on several use cases in different CEA centers.
This post-doc offer concerns the implementation of the CEA/DES (Energy Department) demonstrator on the use case of laser cutting under constraints for A&D at the Simulation and Dismantling Techniques Laboratory (LSTD) at the CEA Marcoule.