This PhD thesis is concerned with cracking in nuclear fuels at the microstructure level, a phenomenon that is essential to understand in order to model the behavior of materials under irradiation. Indeed, crack initiation and propagation can lead to the release of fission gases and the formation of fragments inducing fissile matter displacement. Current industrials models are based on simplified representations of the porous microstructure and empirical fracture criteria, which limits their physical accuracy and validation by separate effects.

To overcome these limitations, the proposed thesis work consists of using multi-scale approaches and high-performance computing (HPC) finite element simulations. The main objectives are to define a Representative Volume Element (RVE) for crack initiation in materials with random porosity, improve the failure criteria used in legacy codes and define their uncertainties, and finally establish the domain of validity for analyzing crack propagation in the RVE.

The first line of research consists of rigorously defining the size of the RVE based on local physical variables such as the maximum principal stress. Variance reduction methods will be used to optimize the number of calculations required and estimate the associated errors.

In a second step, simulations performed to determine the RVE size will be used to improve industrial models. The approach will seek to separate the mechanical effects of an isolated bubble from those resulting from interactions between neighboring bubbles. Machine learning techniques may be used to develop this new model. Validation will be based on indirect measurements of cracking, such as gas release observed during thermal annealing, particularly for high burn-up structure (HBS) fuels, where legacy models fail to predict the kinetics of cracking.

Finally, crack propagation within the RVE will be studied using 3D phase field simulations, which allow for detailed representation of the various stages after the crack initiation. The influence of boundary conditions on the RVE will be examined by comparison with simulations on larger domains.

The thesis will be carried out at the Institute for Research on Nuclear Systems for Low-Carbon Energy Production (IRESNE) of the CEA Cadarache, within the PLEIADES platform development team, which is specialized in fuel behavior simulation and multiscale numerical methods. It will be conducted in collaboration with the CNRS/LMA as part of the MISTRAL joint laboratory, notably on aspects relating to the analysis of random medium representativeness and micromechanical simulation of crack propagation.