About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Characterization of Minerals, Metals and Materials 2026 - In-Situ Characterization Techniques
|
| Presentation Title |
Machine Learning Assisted Serial Sectioning to Enable Rapid 3D Crack Network Reconstruction and Analysis |
| Author(s) |
Gregory B. Thompson, Alyssa Stubbers, Ben Swartley, Sierra Durkee, Evan Schwind, Edgar Solano , Alejandro Ramirez, Christopher R Weinberger, Olivia Graeve, Mireya Sarai Garcia |
| On-Site Speaker (Planned) |
Gregory B. Thompson |
| Abstract Scope |
Serial sectioning techniques enable the 3D reconstruction of microstructures, providing a means of detailed characterization and unique perspectives on phase and defect formations in materials. However, collecting serial section data is time-consuming because of the large number of images required to create a reliable reconstruction. This talk will describe serial sectioning at various cut widths between slices with convolutional neural network (CNN) approaches to interpolate crack networks in ζ-Ta4C3-x. Three crack types (planar, bifurcating, and kinking) were identified to evaluate the specific capacity of the CNN to recreate the complex ζ-Ta4C3-x cracking environment. The accuracy of the CNN-assisted serial sectioning datasets was evaluated using direct image comparison and analysis of the crack surface area. The presented results demonstrate that serial sectioning time could be decreased by 90% with < 1% total image error. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Ceramics, Characterization, Machine Learning |