| About this Abstract |
| Meeting |
TMS Specialty Congress 2025
|
| Symposium
|
The 7th International Congress on 3D Materials Science (3DMS 2025)
|
| Presentation Title |
Deep Learning Enabled Rapid 3D X-ray Tomography for In Situ Mechanical Characterization |
| Author(s) |
Nathan S. Johnson, Hrishikesh Bale, Steve Kelly, Newell Moser, Orion Kafka |
| On-Site Speaker (Planned) |
Nathan S. Johnson |
| Abstract Scope |
Laboratory X-ray characterization techniques often face limitations due to lower flux and brightness compared to synchrotron sources, leading to extended counting times for acquiring high-quality images. This issue is particularly acute for in situ experiments that require multiple datasets, sometimes stretching data collection to several days per sample. In this study, we introduce a deep learning-based reconstruction method tailored for sparse X-ray microscopy datasets. During an in situ tensile test on additively manufactured Inconel 718 dogbone specimens, we collected 1,000 two-dimensional projections over the course of an hour. Utilizing a commercial deep learning model (Zeiss DeepRecon) we achieved high-resolution 3D reconstructions using 100 projections, reducing acquisition time by 10X. The model captured up to 80% of pores and preserved the morphology of the largest pores. This reduction—from one hour to six minutes — notably enhances the efficiency of in situ experiments. |
| Proceedings Inclusion? |
Undecided |