Development of automatic gamma spectrum analysis using a hybrid machine learning algorithm for the radiological characterization of nuclear facilities decommissioning.

The application of gamma spectrometry to radiological characterization in nuclear facility decommissioning, requires the development of specific algorithms for automatic gamma spectrum analysis. In particular, the classification of concrete waste according to its level of contamination, is a crucial issue for controlling decommissioning costs.
Within CEA/List, LNHB, in collaboration with CEA/DEDIP, has been involved for several years in the development of tools for the automatic analysis of low-statistics gamma spectra, which can be applied to scintillator detectors (NaI(Tl), plastics). In this context, an original approach based on a hybrid machine learning/statistics spectral unmixing algorithm has been developed for the identification and quantification of radionuclides in the presence of significant deformations in the measured spectrum, due in particular to interactions between the gamma emission from the radioactive source and its environment.
The proposed subject follows on from thesis work that led to the development of the hybrid algorithm with the aim of extending this approach to the radiological characterization of concrete surfaces. The candidate will be involved in the evolution of the hybrid machine learning/statistical algorithm for the characterization of concrete for classification as conventional waste. The work will include a feasibility study of modeling the deviations of the learned model to optimize the robustness of decision-making.

Novel architecture and signal processing for mobile optical telecommunications

Free-Space Optical Communications (FSO) rely on transmitting data via light between two distant points, eliminating the need for fibers or cables. This approach is particularly valuable when wired connections are impractical or prohibitively expensive.
However, these links are highly susceptible to atmospheric conditions—fog, rain, dust, and thermal turbulence—which attenuate or distort the light beam, significantly degrading communication quality. Current solutions remain costly and limited, both in terms of optical compensation hardware and signal processing algorithms.

Within this framework, the thesis aims to design high-performance, robust mobile optical links capable of adapting to dynamic and disturbed environments. The study will focus on leveraging Silicon-based Optical Phased Arrays (OPAs)—a technology derived from low-cost LiDAR systems—offering a promising path toward compact, integrated, and cost-effective architectures.
The primary focus of the research will be developing advanced algorithmic approaches for signal processing and compensation. The PhD candidate will be tasked with designing a dedicated simulation environment to evaluate and validate architectural choices and algorithmic strategies before practical experimentation.

The overarching goal is to propose an integrated, flexible, and reliable architecture that ensures uninterrupted optical communication in motion, with potential applications in aerospace, space, and terrestrial domains.

Investigation of polytopal methods apllied to CFD and optimized on GPU architecture

This research proposal focuses on the study and implementation of polytopal methods for solving the equations of fluid mechanics. These methods aim to handle the most general meshes possible, overcoming geometric constraints or those inherited from CAD operations such as extrusions or assemblies that introduce non-conformities. This work also falls within the scope of high-performance computing, addressing the increase in computational resources and, in particular, the development of massively parallel computing on GPUs.

The objective of this thesis is to build upon existing polytopal methods already implemented in the TRUST software, specifically the Compatible Discrete Operator (CDO) and Discontinuous Galerkin (DG) methods. The study will be extended to include convection operators and will investigate other methods from the literature, such as Hybrid High Order (HHO), Hybridizable Discontinuous Galerkin (HDG), and Virtual Element Method (VEM).

The main goals are to evaluate:
1. The numerical behavior of these different methods on the Stokes/Navier-Stokes equations;
2. The adaptability of these methods to heterogeneous architectures such as GPUs.

Robust multi-material topological optimization under manufacturability constraints applied to the design of superconducting magnets for high-field MRI

MRI scanners are invaluable tools for medicine and research, whose operation is based on exploiting the properties of atomic nuclei immersed in a very intense static magnetic field. In almost all MRI scanners, this field is generated by a superconducting electromagnet.

The design of electromagnets for MRI must meet very demanding requirements in terms of the homogeneity of the field produced. In addition, as the magnetic field becomes more intense, the forces exerted on the electromagnet increase, raising the issue of the mechanical strength of the windings. Finally, the “manufacturability” of the electromagnet imposes constraints on the shapes of acceptable solutions. The design of superconducting electromagnets for MRI therefore requires a meticulous effort to optimize the design, subject to constraints based on magneto-mechanical multiphysics modeling.

A new innovative multiphysics topological optimization methodology has been developed, based on a density method (SIMP) and a finite element code. This has made it possible to produce magnet designs that meet the constraints on the homogeneity of the magnetic field produced and on the mechanical strength of the windings. However, the solutions obtained are not feasible in practice, both in terms of the manufacturability of the coils (cable windings) and their integration with a supporting structure (coils held in place by a steel structure).

The objective of this thesis is to enhance the topological optimization method by formalizing and implementing manufacturing constraints related to the winding method, residual stresses resulting from pre-tensioning the cables during winding, and the presence of a structural material capable of absorbing the forces transmitted by the coils.

Advanced methods of blockwise diffusion imaging for studying fetal cerebral development at the mesoscopic scale

The second half of pregnancy is an extremely rich period in terms of brain development, during which key processes such as neurogenesis, neuronal migration, and axonal growth take place; transient structures form and disappear, while brain volume increases more than tenfold. A blockwise ex-vivo imaging technique recently developed in NeuroSpin allows us to take a new look on developing brain tissues, leveraging ultra-high-field MRI at 11.7 teslas to acquire unprecedented whole-brain images at mesoscopic resolution (100 to 200 µm 3D isotropic) . The acquired data is highly multiparametric, including quantitative T1, T2, and T2* mapping, as well as high angular resolution, multi-shell diffusion-weighted imaging (b = 1500, 4500, 8000 s/mm² with 25, 60, and 90 directions respectively) at 200 µm isotropic resolution.
In order to reach such a high level of detail, a small-bore scanner is used (5 cm usable diameter) over extended scanning times (150 hours per field of view). Brains older than about 20 gestational weeks are too large, and are sectioned into blocks whose size is compatible with the scanner. The resulting blockwise images are registered using a dedicated semi-automatic protocol, and fused to reconstruct a set of whole-brain images. While this protocol has allowed us to obtain good-quality images on several fetal brain specimens (3 published, 3 other brains in progress as of the end of 2025), the diffusion imaging data remains to be fully analyzed: indeed, the blockwise nature of the acquisitions poses unique challenges, notably due to the discontinuity at the boundary between blocks, but also to non-linear image deformations and non-linearity of the magnetic field gradients.
The PhD candidate will be hosted in the inDEV team (imaging neurodevelopmental phenotypes) in close collaboration (co-supervision) with the Ginkgo team, which has leading expertise in diffusion imaging methods and has pioneered the blockwise acquisition technique in an adult brain known as Chenonceau. The PhD work lies at the interface between imaging, algorithmics, and developmental neuroscience: it will include developing and benchmarking new methods for processing this blockwise diffusion MRI to obtain high-quality tractography and fit diffusion microstructural models. It will also include an experimental part, where the PhD candidate will take part in the acquisition and reconstruction of new brains, both typical specimens and pathological ones with agenesis of the corpus callosum. Finally, the candidate will explore neuroscientific outcomes of this unprecedented dataset, which has exceptional potential to describe processes such as the development of subcortical pathways and associative white matter fibre tracts, and to become the first atlas of the developing fetal brain with fibre architecture at the mesoscopic scale.

Reduction of reinforcement in reinforced concrete structures through nonlinear calculations and topological and evolutionary optimizations

Reinforcing steel plays a major role in the behavior of reinforced concrete structures. Nevertheless, significant conservatisms may sometimes be imposed by design codes, raising questions about the feasibility of construction or the viability of the structure (economic, environmental, etc.). It is within this context that the doctoral research takes place. Building on recent developments, the work aims to propose an innovative design approach relying on the use of nonlinear finite element calculations, combined with topological optimization algorithms (defining reinforcement directions and bar cross-sections) and evolutionary optimization algorithms (determining the placement of bars with fixed cross-sections).
The method should, through an iterative process, yield solutions that meet an optimal design configuration. Considering the multiple, potentially conflicting objectives to minimize (such as cost, feasibility, strength, and carbon footprint), the approach will guide the configuration of input parameters based on an analysis of the relevant output results.
Applying the method to complex, practice-based case studies (for example, beam-column junctions) will demonstrate its relevance compared with more conventional design methods. By the end of the thesis, the doctoral candidate will have developed advanced skills in the use and development of state-of-the-art tools, ranging from nonlinear finite element simulation to modern optimization techniques based on artificial intelligence.

High-Fidelity Monte Carlo Simulations of Neutron Noise in Nuclear Power Reactors

Operating nuclear reactors are subject to a variety of perturbations. These can include vibrations of the fuel pins and fuel assemblies due to fluid-structure interactions with the moderator, or even vibrations of the core barrel, baffle, and pressure vessel. All of these perturbations can lead to small periodic fluctuations in the reactor power about the stable average power level. These power fluctuations are referred to as “neutron noise”. Being able to simulate different types of in-core perturbations allows reactor designers and operators to predict how the neutron flux could behave in the presence of such perturbations. In recent years, many different research groups have worked to develop computational models to simulate these sources of neutron noise, and their resulting effects on the neutron flux in the reactor. The primary objective of this PhD thesis will be to bring Monte Carlo neutron noise simulations to the scale of real-world industry calculations of nuclear reactor cores, with a high-fidelity continuous-energy physics representation. As part of this process, the student will add novel neutron noise simulation capabilities to TRIPOLI-5, the next-generation production Monte Carlo particle-transport code jointly developed by CEA and ASNR, with the support of EDF.

Preconditioning of iterative schemes for the mixed finite element solution of an eigenvalue problem applied to neutronics

Neutronics is the study of the behavior of neutrons in matter and the reactions they induce, particularly the generation of power through the fission of heavy nuclei. Modeling the steady-state neutron flux in a reactor core relies on solving a generalized eigenvalue problem of the form:
Find (phi, keff) such that A phi=1/keff B phi and keff is the eigenvalue with the largest magnitude, where A is the disappearance matrix which is assumed invertible, B represents the production matrix, phi denotes the neutron flux, and keff is called the multiplication factor.

The neutronics code APOLLO3® is a joint project of CEA, Framatome, and EDF for the development of a next-generation code for reactor core physics to meet both R&D and industrial application needs [4].
The MINOS solver [2] is developed within the framework of the APOLLO3® project. This solver is based on the mixed finite element discretization of the neutron diffusion model or the simplified transport model. The strategy for solving the aforementioned generalized eigenvalue problem is iterative; it involves applying the inverse power method [6].

The convergence speed of this inverse power method algorithm depends on the spectral gap. In the context of large cores such as the EPR reactor, it is observed that the spectral gap is close to 1, which degrades the convergence of the inverse power method algorithm. It is necessary to apply acceleration techniques to reduce the number of iterations [7]. In neutron transport, the preconditioning called Diffusion Synthetic Acceleration is very popular for the so-called inner iteration [1] but has also recently been applied to the so-called outer iteration [3]. A variant of this method was introduced in [5] for solving a source problem. It is theoretically shown that this variant converges in all physical regimes.

[1] M. L. Adams, E. W. Larsen, Fast iterative methods for discrete-ordinates particle transport calculations, Progress in Nuclear Energy, Volume 40, Issue 1, 2002.

[2] A.-M. Baudron and J.-J. Lautard. MINOS: a simplified PN solver for core calculation. Nuclear Science and Engineering, volume 155(2), pp. 250–263 (2007).

[3] A. Calloo, R. Le Tellier, D. Couyras, Anderson acceleration and linear diffusion for accelerating the k-eigenvalue problem for the transport equation, Annals of Nuclear Energy, Volume 180, 2023.

[4] P. Mosca, L. Bourhrara, A. Calloo, A. Gammicchia, F. Goubioud, L. Mao, F. Madiot, F. Malouch, E. Masiello, F. Moreau, S. Santandrea, D. Sciannandrone, I. Zmijarevic, E. Y. Garcia-Cervantes, G. Valocchi, J. F. Vidal, F. Damian, P. Laurent, A. Willien, A. Brighenti, L. Graziano, and B. Vezzoni. APOLLO3®: Overview of the New Code Capabilities for Reactor Physics Analysis. Nuclear Science and Engineering, 2024.

[5] O. Palii, M. Schlottbom, On a convergent DSA preconditioned source iteration for a DGFEM method for radiative transfer, Computers & Mathematics with Applications, Volume 79, Issue 12, 2020.

[6] Y. Saad. Numerical methods for large eigenvalue problems: revised edition. Society for Industrial and Applied Mathematics, 2011.

[7] J. Willert, H. Park, and D. A. Knoll. A comparison of acceleration methods for solving the neutron transport k-eigenvalue problem. Journal of Computational Physics, 2014, vol. 274, p. 681-694.

Study of the behaviour of mixed oxide fuels with degrade isotopy at the beginning of life.

France has decided to adopt a 'closed' nuclear fuel cycle. This involves processing spent fuel to recover valuable materials such as uranium and plutonium, while other compounds such as fission products and minor actinides constitute final waste. UO2 fuel irradiated in pressurised water reactors (PWRs) is currently reprocessed to produce plutonium (PuO2), which is then reused in the form of mixed oxide (MOX) fuel. This fuel is then irradiated in PWRs, a process known as plutonium monorecycling. The CEA is currently studying the multi-recycling of materials using fuels containing Pu from the processing of spent MOX assemblies. However, this multi-recycled plutonium contains a higher proportion of highly alpha-active isotopes (Pu238, Pu240 and Pu241/Am241), resulting in more severe alpha self-irradiation than current MOX fuels experience [1]. This exacerbates certain physical phenomena [2-5], such as fuel swelling due to helium precipitation and the creation of crystal defects and decreased thermal conductivity [6-8], which can alter its behaviour in the reactor.
The proposed thesis will study the impact of these phenomena on the behaviour of MOX fuels at the beginning of the irradiation, using a combination of experimentation and modelling. Heat treatments will be employed to analyse the mechanisms of crystal defect healing and helium behaviour. Various experimental techniques will be employed to characterise the structure and microstructure (X-ray diffraction, scanning electron microscopy (SEM), Raman spectroscopy and microprobe analysis), defect densities (transmission electron microscopy (TEM)), helium release (KEMS), thermal gradient reproduction (CLASH laser) and thermal conductivity (LAF laser). The results will inform simulations modelling the microstructure and thermal properties.
This cross-disciplinary study will improve our understanding of the phenomena involved in the initial power-up of fuels damaged by alpha self-irradiation, particularly the impact of helium produced by decay.

You will be based at the Multi-Fuel Design and Irradiation Laboratory (LECIM) within the Research Institute for Nuclear Systems for Low-Carbon Energy Production at CEA/Cadarache. For the experimental part of the project, you will collaborate with the Chemical Analysis and Materials Characterisation Laboratory (LMAT) at CEA/Marcoule and the European Research Centre (JRC) in Karlsruhe. You will have the opportunity to publish your results through scientific publications and conference presentations. This role offers the chance to develop your expertise in a variety of techniques that can be applied across multiple fields of materials science and engineering.

[1]O. Kahraman, thésis, 2023.[2]M. Kato et al., J Nucl Mater, 393 (2009) 134–140.[3]L. Cognini et al., Nuclear Engineering and Design 340 (2018) 240–244.[4] T. Wiss et al., Journal of Materials Research 30 (2015) 1544–1554.[5]D. Staicu et al., J Nucl Mater 397 (2010) 8–18.[6] T. Wiss et al.,Front. Nucl. Eng. 4 (2025) 1495360.[7]E.P. Wigner, J. Appl. Phys. 17 (1946) 857–863.[8]D. Staicu et al., Nuclear Materials and Energy 3–4 (2015) 6–11.

From optimal control in NMR at 11.7 Tesla to precision imaging of the human brain in vivo