About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
Presentation Title |
Improving Prediction of Microstructures Using Physics-informed Machine Learning |
Author(s) |
Joseph Hafen, Benjamin Rhoads, Samrat Choudhury |
On-Site Speaker (Planned) |
Benjamin Rhoads |
Abstract Scope |
Phase field simulations can predict the formation of microstructure in materials under a given set of external processing conditions. However, these simulations require significant computational resources, especially when three dimensional microstructures are desired. To expedite the simulation process, machine learning tools can be tailored to make data-driven microstructure predictions, requiring far less computational resources in comparison to phase field simulation. This work feeds phase field generated microstructures of oxide-based ferroelectric material through a trained autoencoder. To improve the performance of this autoencoder, the energetic components that drives the formation of the microstructure are calculated and used to improve the accuracy of the autoencoder predictions. Alternatively, with this additional information in the model, the autoencoder is able to retain similar accuracy in its predictions while using significantly less phase field simulated microstructures, dramatically reducing the computational resources required. |
Proceedings Inclusion? |
Planned: |
Keywords |
Computational Materials Science & Engineering, Machine Learning, |