About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
H-12: High-Throughput, High-Fidelity Characterization of Complex Alloys via Machine Learning Driven 4D-STEM |
| Author(s) |
Daniel A. Bolden, Louis Katz, Ravit Silverstein, Arda Genc |
| On-Site Speaker (Planned) |
Daniel A. Bolden |
| Abstract Scope |
Multidimensional transmission electron microscopy (TEM), particularly four-dimensional scanning TEM (4D-STEM), is revolutionizing high-resolution materials characterization. This progress is driven by fast, pixelated detectors that record diffraction patterns at each probe position, allowing for the detection of subtle structural changes with sub-angstrom spatial resolution. However, efficiently processing these large datasets (> 10 GB) with high accuracy remains a challenge.Here, we present an end-to-end, high-speed (~300 frames/s) processing framework built on a sub-pixel-accurate, neural network–based object detection algorithm to accelerate data analysis for real-time feedback during experiments in crystallographic phase identification and strain mapping (with 5×10⁻⁴ peak localization precision) of complex, multiphase metallic systems. We demonstrate this method on Ti-Nb alloys, which display intricate misfit strain fields and intermediate phases, thereby setting a benchmark for analyzing candidate multi-principal element alloys. This approach allows high-precision quantitative analysis of short-range order, multidimensional defects, and misfit strain, offering insights for designing high-performance materials. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
High-Entropy Alloys, Machine Learning, Characterization |