| About this Abstract |
| Meeting |
MS&T22: Materials Science & Technology
|
| Symposium
|
Advanced Characterization of Materials for Nuclear, Radiation, and Extreme Environments III
|
| Presentation Title |
Machine Learning Algorithms for High-throughput Characterization of Structure and Microstructure of Metals for Extreme Environments |
| Author(s) |
Nishan Senanayake, Thaddeus Rahn, Nathaniel k Tomczak, Assel Aitkaliyeva, Jennifer LW Carter |
| On-Site Speaker (Planned) |
Jennifer LW Carter |
| Abstract Scope |
A major bottleneck in the nuclear materials research/design space is the near-ubiquitous requirement that gray-scale micrographs require manual interpretation of microstructural features. Conventional image segmentation algorithms often require manual intervention, making it difficult to generalize from one challenge to another. In this work, we present trained machine learning (ML) algorithms for high-throughput 1) classification of phases from selected-area diffraction patterns in the Pu-Zr system and 2) segmentation of secondary electron micrographs in 𝛾′′, 𝛾′ strengthened nickel-based superalloys. The Pu-Zr algorithm utilizes a convolutional neural network (CNN), without integration of materials domain knowledge, to index between the α and δ phases. The proof-of-concept model indicates that full automation of the diffraction pipeline is attainable. The classification of 𝛾′′ and 𝛾′ pixels shows the ML can either increase (random forest, CNN) or decrease (support vector machine) computational efficiency compared to conventional image segmentation without negatively impacting accuracy. |