Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 4: Data Driven Material Science: Big Data and Workflows I
MM 4.8: Vortrag
Montag, 18. März 2024, 12:15–12:30, C 243
Stable diffusion based microstructure reconstruction and generation — •Yixuan Zhang1, Teng Long2, and Hongbin Zhang1 — 1Institute of Materials Science, Technical University of Darmstadt, 64287, Darmstadt, Germany — 2School of Materials Science and Engineering, Shandong University, 250061, Jinan, China
In recent years, the reconstruction and generation of microstructures have become pivotal in understanding and predicting the mechanical and functional properties of materials. This study introduces a novel approach to microstructure reconstruction based on stable diffusion models. Our implementation employs a stable diffusion model to capture the intricate patterns and features inherent in microstructures, which can be adapted to further refine reconstructed the phase and grain orientation of microstructures, ensuring their statistical and morphological fidelity to the original samples. The model is trained using a comprehensive dataset of 500,000 synthetic micrographs, ensuring the model's robustness and versatility across various material classes. Our results demonstrate that our approach outperforms conventional methods in terms of accuracy, speed, and adaptability. The reconstructed microstructures exhibit remarkable similarity to their counterparts, both qualitatively and quantitatively. Furthermore, the generative capabilities of our model pave the way for optimizing novel microstructures, aiding in the design of materials with desired properties.
Keywords: MCR; Stable diffusion; Phase; Grain orientation