Abstract Scope |
Changes in the Grain Boundary Network (GBN) during grain growth significantly impact the microstructure's final properties. Although several studies have employed simulations to investigate microstructure evolution, these simulations can be computationally expensive, particularly when considering anisotropic GB properties. In this work, we propose an alternative approach for obtaining an evolved microstructure without resorting to extensive simulations. Instead, by using a comprehensive dataset comprising thousands of microstructures obtained from anisotropic grain growth simulations,
we reconstruct the morphology of the microstructure using a diffusion model trained on the dataset images. Subsequently, we develop a predictive model based on spectral graph theory to estimate the statistical descriptors and texture of the evolved microstructure. Finally, we compare the reconstructed microstructures with the actual final microstructures obtained from simulations, focusing on achieving close agreement in terms of statistical descriptors. This approach presents a promising avenue for efficiently predicting the evolution of microstructures during grain growth. |