DRAGen - A deep learning supported RVE generator framework for complex microstructure models.

In this study an improved version of the Discrete RVE Automation and Generation Framework, also called DRAGen, is presented. The Framework incorporates a generator for Representative Volume Elements (RVEs). Several complex microstructure features, extracted from real microstructures, have been added to the generator, to enable it to generate RVEs with realistic microstructures. DRAGen is now capable of reading trained neural networks as well as .csv-files as input data for the microstructure generation. Furthermore, features such as pores and inclusions, martensite bands, hierarchical substructures, and crystallographic textures can be reconstructed in the RVEs. Besides the features, the functionality for different solvers was introduced. Therefore, the code was extended by modules for the generation of Finite Element (FE) and spectral solver input files. DRAGen now has the ability to create models for three powerful multiphysics frameworks used in the community: DAMASK, Abaqus and MOOSE. The evaluation of the features, as well as the simulations performed on sample models, show that the new version of DRAGen is a very powerful tool with flexible applicability for scientists in the ICME community. Also, due to the modular architecture of the project, the code can easily be expanded with features of interest. Therefore, it delivers a variety of functions and possible outputs, which offers researchers a broad spectrum of microstructures that can be used in microstructure studies or microstructure design developments.

Generating Input Data for Microstructure Modelling: A Deep Learning Approach Using Generative Adversarial Networks.

For the generation of representative volume elements a statistical description of the relevant parameters is necessary. These parameters usually describe the geometric structure of a single grain. Commonly, parameters like area, aspect ratio, and slope of the grain axis relative to the rolling direction are applied. However, usually simple distribution functions like log normal or gamma distribution are used. Yet, these do not take the interdependencies between the microstructural parameters into account. To fully describe any metallic microstructure though, these interdependencies between the singular parameters need to be accounted for. To accomplish this representation, a machine learning approach was applied in this study. By implementing a Wasserstein generative adversarial network, the distribution, as well as the interdependencies could accurately be described. A validation scheme was applied to verify the excellent match between microstructure input data and synthetically generated output data.