Accelerated Materials Simulations with Deep Neural Networks

Fei Zhou | 22-ERD-016

Executive Summary

Large-scale molecular dynamic simulations are an essential and ubiquitous tool to accelerate the development and deployment of materials relevant to national security missions, but they are too slow and too computationally expensive to adequately meet next-generation mission needs. We will develop and apply machine-learning based models to accelerate large-scale simulations of the evolution of material microstructures and dislocation lines, offering dramatically faster simulation capabilities with unprecedented accuracy.

Publications, Presentations, and Patents

Bertin, Nicolas, and Fei Zhou. 2023. “Accelerating Discrete Dislocation Dynamics Simulations with Graph Neural Networks.” Journal of Computational Physics 487 (April): 112180.

Fei Zhou. 2022. “Accelerate microstructure simulation with machine learning” (Presentation, Artificial Intelligence for Materials Science (AIMS) NIST Conference, Gaiithersburg, MD, Jul 12-14, 2022).

Fei Zhou, “How to take large timesteps in MD? Score dynamics from diffusion model back to Langevin dynamics” (Presentation, MRS Spring Meeting, San Francisco, CA, April 2023).

Peter Park, “Accurate simulation of precipitation events using diffusion probabilistic models” (Presentation, Materials Research Society-MRS Spring Meeting, San Francisco, CA, April 2023).

Tim Hsu, “Accelerated, probabilistic molecular dynamics over picosecond timesteps via diffusion model” (Presentation, MRS Spring Meeting, San Francisco, CA, April 2023).

Fei Zhou, “Probabilistic approach to materials modeling” (Presentation, TMS Annual Meeting, San Diego, CA, March 2023).

Fei Zhou, “Probabilistic approach to materials modeling” (Presentation, AI/ML workshop with CEA-DAM, Paris, France, virtual, June 2023).

Fei Zhou, “Generative AI for microstructure modeling: from processing-structure relationship to mesoscale simulation” (Presentation, UC Davis, Davis, CA, Sept. 21, 2023).

Fei Zhou, “Identifying and Quantifying Uncertainty in Fitted Interatomic Potentials for MD” (Presentation, Materials + Data Science Workshop, University of Minnesota, Minneapolis, MN, Sept. 22, 2023).