Deep Learning Based X-ray Microscopy Reconstruction for Large Volume Correlative Characterization

Authors:

Abstract
Correlative characterization often begins with non-destructive evaluation, such as X-ray computed tomography (XCT) and X-ray microscopy (XRM). Regions or features of interest are identified in the 3D X-ray datasets and then, after identification, are characterized in other, destructive imaging modalities. In the realm of modern X-ray imaging, a critical challenge arises: the pursuit of sub-micron-level defect detection within large volumes. Countless material applications demand statistically relevant samplings of defects, encompassing secondary phases, inclusions, phase segregation, cracks, pores, delamination, and grain structures. However, there is an inherent tradeoff between the volume that can be imaged and the resolution of the images. Imaging at a large volume may yield a statistically relevant sample size but often suffers from poor resolution, rendering it inadequate for detecting these elusive defects. Conversely, high-resolution imaging captures intricate details but confines the examination to a small volume, raising questions about statistical representativeness.