| Abstract Scope |
Grain growth (GG) significantly affects the mechanical properties of materials, making its prediction important in materials science. Traditional methods, such as ToRealMotion (TRM), Monte-Carlo Potts, Cellular Automaton, Vertex, Phase Field, and Level-Set methods, are effective but computationally demanding. This has led to growing interest in more efficient approaches, like machine learning, particularly deep neural networks. Deep learning, especially LSTMs, is well-suited for capturing sequential changes in evolving microstructures and modeling complex patterns over time. Building on this, we developed an LSTM-based model designed for isotropic grain growth. A large database of validated simulations of normal grain growth in an idealized system was created, using diverse parameters to ensure a versatile predictive model. The trained model showed strong agreement with analytical predictions across various test cases. By fine-tuning parameters and training duration, we identified the dataset requirements for accurate and efficient grain growth predictions, improving computational tools in materials science. |