Generative adversarial network (GAN) enabled Statistically equivalent virtual microstructures (SEVM) for modeling cold spray formed bimodal polycrystals

Abstract

AbstractImage-based micromechanical models, necessary for the development of structure-property-response relations, are far from mature for complex microstructures with multi-modal distributions of morphological and crystallographic features, such as those occurring with cold spray-formed (CSF) aluminum alloys. These materials have a bimodal polycrystalline microstructure composed of recrystallized ultra-fine grains (UFGs) and deformed coarse grains (CGs) within prior particles. A prime reason is the lack of robust approaches for generating statistically equivalent virtual microstructures (SEVM) capturing the statistics of characteristic morphological and crystallographic features, such as grain size, crystallographic orientations, and misorientations. This paper introduces an approach, strategically integrating Generative Adversarial Network-based approaches for producing bimodal CSF AA7050 alloy microstructures, with the synthetic microstructure builder Dream3D for packing prior particles with CGs having statistically equivalent morphological and crystallographic descriptors to electron backscatter diffraction (EBSD) maps. An efficient finite element (FE) simulation approach is developed for the SEVMs to generate local and overall response functions through the creation of sub-volume elements (SVEs).