About this Abstract |
Meeting |
MS&T22: Materials Science & Technology
|
Symposium
|
Grain Boundaries, Interfaces, and Surfaces: Fundamental Structure-Property-Performance Relationships
|
Presentation Title |
Simulate Grain Growth with Machine Learning Techniques |
Author(s) |
Shaoxun Fan, Ming Tang, Fei Zhou |
On-Site Speaker (Planned) |
Ming Tang |
Abstract Scope |
Grain growth is a type of microstructure evolution important for many engineering materials. Grain growth is traditionally simulated by phase-field, surface evolver and lattice Monte Carlo models. Here we demonstrate the ability of machine learning (ML) techniques to learn the grain growth rules and predict the phenomenon in both 2D and 3D. Two types of ML models based on convolutional recurrent and graphic neural networks are trained on short image sequences generated by phase-field simulation. Properly trained ML models can extend predictions in temporal and spatial domains by >10 times of the training dataset. In addition to very good pixelwise agreement at short times, the models accurately capture the grain growth statistics in the long term. Compared with phase-field models, ML models accelerate simulations by 2-3 orders of magnitude, and can also forecast the evolution of systems with unknown parameters. |