|About this Abstract
||2022 TMS Annual Meeting & Exhibition
||Advanced Real Time Imaging
||Analysis of In-situ X-ray Tomography Datasets of Dendritic Solidification Using 2D and 3D Machine Learning Algorithms
||Tiberiu Stan, Nathan Pruyne, Jim James, Marcus Schwarting, Jiwon Yeom, Ben Blaiszik, Ian Foster, Peter Voorhees
|On-Site Speaker (Planned)
Many advanced in-situ characterization techniques output large multimodal datasets that must be quantitatively processed to extract materials information. Image segmentation (the act of grouping the pixels of an image into useful parts) is at times the most time-consuming and subjective step in the analysis workflow. We show that it is possible to segment in-situ x-ray computed tomography datasets of dendritic solidification using a variety of 2D and 3D machine learning algorithms. The segmentations are compared both qualitatively (through visual inspection) and quantitatively using three computed metrics: pixel classification accuracy, intersection over union, and boundary F1 scores. Machine learning architectures, training techniques, hyperparameter selection, and general guidelines for implementation will be discussed. These advances in processing of large in-situ datasets will accelerate the rate of materials development, design, and discovery.
||Machine Learning, Solidification, Characterization