About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Computational Thermodynamics and Kinetics
|
| Presentation Title |
Malmsgraf: Harnessing Generative Deep Learning Towards a Unified Model for Predictive Metallography |
| Author(s) |
Frank K. Mittel, Taylor D. Sparks |
| On-Site Speaker (Planned) |
Frank K. Mittel |
| Abstract Scope |
Well known deep learning models such as CrabNet have successfully demonstrated the ability to predict the properties of bulk materials based on composition and other related characteristics; however, such models lack spatial context and reasoning which renders them unable to make predictive inference with regards to morphological properties and behavior such as microstructure, phase separation, and domain evolution. We present Malmsfgraf, a universal framework and generative deep learning model that is capable of predicting material properties within the spatial domain from known constraints (e.g. composition and thermal processing conditions) which are transformed by the model into a vectorized form. We will present our research into the interpolative and extrapolative abilities of our model as well as its ability to be trained on new data for expansion into novel regimes of application. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, Other |