| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Integrated Computational Materials Engineering for Physics-Based Machine Learning Models
|
| Presentation Title |
Statistically Equivalent Virtual Microstructures for Modeling of Complex Polycrystalline Alloys Using a Generative Adversarial Network (GAN)-Enabled Computational Platform |
| Author(s) |
Joshua Stickel, Brayan Murgas, Luke Brewer, Somnath Ghosh |
| On-Site Speaker (Planned) |
Joshua Stickel |
| Abstract Scope |
This presentation introduces a methodology for creating 2D image based, 3D statistically equivalent virtual microstructures (SEVMs) for polycrystalline materials with complex microstructures encompassing multi-modal morphological and crystallographic distributions. Cold spray formed (CSF) Al 7050 alloys, containing prior particles with coarse grains (CGs) and ultra-fine grains (UFGs) are one such example of these materials. An integrated SliceGAN-Dream3D platform is introduced consisting of a Generative Adversarial Network (GAN) to reconstruct complex morphology of multi-modal regions, with Dream3D for packing grains into these regions individually conforming to the experimental statistics in EBSD maps. A robust multiscale model is developed coupling crystal plasticity (CP) FEM for coarse-grained polycrystalline microstructures with an upscaled constitutive model (UCM) for modeling UFGs in a self-consistent manner. Finally, the micro-mechanical response is explored by averaging the response of sub-volume elements (SVEs). The SEVM generation and simulation constitute important components of image-based micromechanical modeling, necessary for microstructure-property relations. |