| About this Abstract |
| Meeting |
TMS Specialty Congress 2024
|
| Symposium
|
Accelerating Discovery for Mechanical Behavior of Materials 2024
|
| Presentation Title |
Automated Feature Extraction for Identifying Structure-property Relationships |
| Author(s) |
William Frieden Templeton, Justin Miner, Sneha Prabha Narra |
| On-Site Speaker (Planned) |
Sneha Prabha Narra |
| Abstract Scope |
Microstructural analysis is essential for establishing process-structure-property relationships in new materials and processes. However, the quantitative representation of sparse, complex, and spatially intertwined microstructural features in micrographs poses a challenge. In this work, we employ a variational autoencoder framework to encode micrographs into a latent space with minimal information loss. By analyzing the latent representation, we can identify certain dimensions that correlate with our property of interest, thus linking encoded features from micrographs to key properties. Further, visualizing these dimensions through the inverse transform enhances the interpretability of our analysis and enables us to understand the features that govern the property. Our results provide initial insights into the potential versatility of this approach for automated extraction of informative microstructural features that could help determine structure-property relationships. |
| Proceedings Inclusion? |
Definite: Other |