Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 33: Machine Learning in Dynamics and Statistical Physics I
DY 33.7: Talk
Thursday, March 20, 2025, 11:00–11:15, H47
stable diffusion for microstructure: from microstructural properties to 2D-to-3D reconstruction — •Yixuan Zhang1, Teng Long2, Mian Dai1, and Hongbin Zhang1 — 1TU Darmstadt, Darmstadt, Germany — 2Shandong University, Jinan, China
We propose a novel framework that combines Stable Diffusion and ControlNet to generate microstructures tailored to specific properties, such as coercivity. By leveraging latent alignment techniques, our method enables direct reconstruction of 3D microstructures from 2D inputs, ensuring geometric and property consistency across dimensions. This approach not only facilitates accurate 2D-to-3D reconstruction but also opens possibilities for studying and predicting microstructural transformations during various manufacturing processes. By integrating generative AI with material design, this work provides a robust foundation for property-driven microstructure generation, offering a potential pathway to optimize materials for targeted applications.
Keywords: stable diffusion; microstructure; property-driven; latent alignment