Chemical heterogeneity revealed by quantitative x-ray imaging

X-ray imaging and absorption spectroscopy provide non-destructive methods to study material structures, revealing composition and chemical states at micrometer to sub-100nm scales. However, data imperfections and noise, along with data complexity, hinder quantitative analysis. To address this, integrating physical models with machine learning is improving accuracy. This presentation highlights our recent work at NSLS-II on 2D and 3D chemical state analysis. We’ll discuss a machine learning method to correct uneven background in 2D XANES from TXM, improving detection sensitivity and analysis accuracy. Next, we’ll present a strategy reducing energy points required in typical XANES measurement, which speeds data collection time by a factor of five. Building on this energy selection scheme, we developed a new neural network-based reconstruction algorithm, enabling efficient 3D XRF-XANES data collection within a single tomography data collection cycle— a stark contrast to traditional measurements that take weeks due to the extensive number of energy points and 2D projections required in the scanning X-ray microscopy.