Just another WordPress site - Ruhr-Universität Bochum

Abstract

A novel strategy is presented to derive 3D stereographic information from 2-D sections by machine learning trained by 3-D phase-field simulations. The simulations contain full information of characteristic microstructural features, such as volume fraction of phases, grain sizes and their distribution, percolation of grains, ruggedness etc. AND information about their evolution in dependence of processing and testing conditions. From 3-D microstructures several hundreds or thousands of 2-D sections can be evaluated and used as training sets to retrieve the 3-D information. This information can be, in a second step connected to the information of microstructure evolution, i.e. to include the process history. The trained machine then can be applied to extract full 3-D information from 2-D experimental microstructures. The strategy is supposed to strengthen significantly the analysis of 2-D microstructures if compared to approaches which are trained by 2-D simulations only. The presentation will first highlight the ability of phase-field simulations for property predictions and demonstrate first steps with the new strategy.