| About this Abstract |
| Meeting |
TMS Specialty Congress 2025
|
| Symposium
|
The 7th International Congress on 3D Materials Science (3DMS 2025)
|
| Presentation Title |
Classification of Abnormal Grain Growth Using 3D Convolutional Neural Networks |
| Author(s) |
Woohyun Eum, Yi Wang, Amanda Krause, Michael Tonks, Joel B. Harley |
| On-Site Speaker (Planned) |
Woohyun Eum |
| Abstract Scope |
Understanding what triggers abnormal grain growth (AGG) remains a challenge, as conventional descriptors do not fully explain it. This paper hypothesizes that more complex descriptors play a role in driving AGG. To test this, we train a 3D convolutional neural networks (CNN) to identify abnormal grains (identified manually after heat treatment) from the microstructure boundaries as-sintered Nickel samples before heat treatment, where human experts cannot visually distinguish abnormal grains. Our results demonstrate that the model achieves a 81.6% accuracy, as compared with a 66% accuracy for classifier based only on grain size. Note our input data does not contain orientation information, implying that abnormal grain growth may be observed from the grain boundary network alone. This work marks the first use of 3D CNNs with 3D measurements to capture multi-dimensional information about microstructures, highlighting their critical role in predicting AGG at early stages, without relying on traditional metrics. |
| Proceedings Inclusion? |
Undecided |