Search

International audience; During the last decades, advances in 3D characterization have been achieved to image the Solid Oxide Cells (SOC) electrode microstructure with a high spatial resolution by using different techniques (FIB-SEM, X-ray tomography…). These techniques enable to study the links between the electrode microstructural and physical properties. However, this approach is time consuming as it requires the manufacturing and the characterization of several cells. An alternative consists in generating representative synthetic microstructures by numerical means in order to increase the amount of data required to establish the correlations linking the electrode microstructure parameters. The proposed talk will then introduce different ways to model and simulate virtual SOC electrode microstructures, of both two-phase electrodes and three-phased composite electrodes, using stochastic geometry. The performance, representativeness and flexibility of such models will be demonstrated and validated on real 3D reconstructions. The author(s) acknowledge(s) the support of the French Agence Nationale de la Recherche (ANR), under grant ANR-18-CE05-0036 (project ECOREVE).

International audience; This paper presents a new method for estimating the parameters of a stochastic geometric model for multiphase flow image processing using local measures. Local measures differ from global measures in that they are only based on a small part of a binary image and consequently provide different information of certain properties such as area and perimeter. Since local measures have been shown to be helpful in estimating the typical grain elongation ratio of a homogeneous Boolean model, the objective of this study was to use these local measures to statistically infer the parameters of a more complex non-Boolean model from a sample of observations. An optimization algorithm is used to minimize a cost function based on the likelihood of a probability density of local measurements. The performance of the model is analysed using numerical experiments and real observations. The errors relative to real images of most of the properties of the model-generated images are less than 2%. The covariance and particle size distribution are also calculated and compared.