Processing uncertain microstructural data

The objective of our research is the image-based prediction of the material properties of materials with stochastic microstructures. The uncertain microstructure information is gathered from dedicated synthetic microstructures as well as from experimental measurements. FFT-accelerated FEM simulations are used to build in silico databases for further processing with unsupervised and supervised machine learning tools, e.g. for the automatic feature extraction and for the development of dedicaetd substitute models. The resulting surrogate models will be developed for predicting the effective material response based on image data only. These predictions will be the basis of further processing in Monte Carlo studies for the quantification of the uncertainty of the response. Cross-validation using new datasets and close interaction with the tandem project PN5-4 contribute to the scientific advance.