| About this Abstract |
| Meeting |
2023 TMS Annual Meeting & Exhibition
|
| Symposium
|
Powder Materials Processing and Fundamental Understanding
|
| Presentation Title |
A-58: Instance Segmentation for the Characterization of Metal Powders Using Synthetic Datasets |
| Author(s) |
Kyle Farmer, Ryan Cohn, Elizabeth Holm |
| On-Site Speaker (Planned) |
Kyle Farmer |
| Abstract Scope |
Powder characteristics can significantly impact the manufacturing and performance of powder based additive manufactured materials. Instance segmentation, using Mask R-CNN with transfer learning, has shown success in characterizing the size distribution of gas atomized powders. However, the manual annotation process to provide ground truth labels for training is expensive and difficult. In this work, we propose training Mask R-CNN with synthetic datasets which will help alleviate the time cost and increase the accuracy of ground truth labels while yielding high accuracy during evaluation of real gas atomized powders. The synthetic datasets consist of eight closely related particle size distributions. The resulting particle size measurements agree closely with their underlying distributions. The model also showed success in predicting the overlapping regions of particles, despite their being hidden from view.
Honeywell Federal Manufacturing & Technologies, LLC operates the Kansas City National Security Campus for the United States Department of Energy / National Nuclear Security Administration under Contract Number DE-NA0002839 NSC-614-4681 dated 07/2022 Unclassified Unlimited Release. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Powder Materials, Characterization |