Reconstruction of Microstructural and Morphological Parameters for RVE Simulations with Machine Learning

Abstract

Advanced High Strength Steels (AHSS) are of great interest for the automotive industry. Since they are applied in a diverse context their properties have to adhere to this diversity. The mechanical properties of modern AHSS are therefore tailored to their specific field of application. Thus they are able to fulfill the desired requirements of the technology. By adapting the steel to a specific area of application, however, these materials obtain a multi-phase microstructure with increasingly complex morphologies. To investigate the influence of these morphological properties on the mechanical properties, Representative Volume Elements (RVE) are commonly used. These RVE are generated based on statistical data and algorithms using randomization methods to get virtual microstructure models that adhere to the real material’s microstructural parameters. This study introduces an RVE generator which is capable of generating statistically representative and periodic RVE. Additionally, the RVE are periodically meshed, so that periodic loading conditions are applicable. Since the input parameters are the most important part of generating the RVE, a machine learning algorithm was trained to reproduce input data equivalent to the real material’s microstructural and morphological parameters.