Séminaire des doctorants Orléans
Titouan Simonnet: Identification and quantification of mineralogical phases in complex 3D materials: use of Artificial Neural Networks in XRD tomographic studiesTitouan Simonnet (IDP-Orléans)
Thursday 17 October 2024 10:30 - IDP-Orléans - Salle de séminaire
Résumé :
Both natural (e.g, soils) and engineered solids are complex and consist of many phases, including minerals. Each of these minerals has its own specific chemical behaviour when in contact with water, and has its own mechanical strength. In addition, the way the different minerals connect within the solid defines the chemical and mechanical properties of the solid. Thus, determining the nature and three-dimensional organisation of the minerals present in a solid is a fundamental requirement for our ability to understand, model and predict its mechanical properties. Synchrotron X-ray tomographic studies are ideally suited for the analysis of these solids. They allow the recording of three-dimensional data with a voxel (3D pixel) at the nanometric scale. These analyses lead to a very large amount of data, so that the processing of these data must be automatic. However, traditional data processing techniques, such as principal component analysis (PCA), are not efficient. The objective of the thesis will be to perform a complete three-dimensional reconstruction of the mineralogy of these solids. This will be achieved by using artificial neural networks (ANNs) on data of increasing complexity. The focus will be on identifying and quantifying mineralogical phase mixing at the voxel scale. In particular, we will develop a study of the uncertainty on the mixing proportions estimated by ANN.
Liens :