Skip to main content
Back to timeline
arXivSource publication:

When diffraction spots blur together, grain-size distributions still come out: hierarchical Bayes separates overlapping peaks, checked against EBSD in Cu-Ti alloy

Synopsis

Measuring grain-size distributions from single-shot X-ray diffraction used to depend on discrete, countable diffraction spots; when many grains or strong texture crowd the spots, counting fails. A hierarchical Bayesian method uses reversible-jump Markov chain Monte Carlo to infer spot number, positions and intensities jointly with a shared intensity-distribution prior, then converts intensity to grain volume. On synthetic patterns where the mean nearest-neighbor spot spacing is about 0.786 times the full width at half maximum and 74.7% of spots have a neighbor within one FWHM, intensity recovery still has a root mean squared error below 0.02; Cu-Ti alloy foil patterns agree more closely with electron backscatter diffraction for the larger-grain specimens.

Source-provided article image: Grain-size distribution characterization from single-shot x-ray diffraction using hierarchical Bayesian multi-emitter analysis
arXiv

Interpretation

A single two-dimensional X-ray diffraction pattern can now yield a grain-size distribution without requiring the diffraction spots to be discrete and countable. Earlier spot-counting routes could not separate individual grain contributions when the scattering volume held many grains or when crystallographic texture crowded spots in some azimuths. On synthetic patterns with mean nearest-neighbor spacing about 0.786 times the FWHM and 74.7% of spots having a neighbor within one FWHM, intensity-distribution recovery had a root mean squared error below 0.02.

The approach places all pixels of a diffraction ring into one joint fit so spots and the intensity distribution constrain each other. Reversible-jump Markov chain Monte Carlo explores plausible combinations of overlapping spots by adding, removing, splitting and merging diffraction components while updating positions and intensities. The intensity prior is a finite mixture of up to three lognormal components, and the component count, means, widths and weights are inferred together with the spot catalog, so no pre-calibrated intensity prior is needed.

The experimental comparison uses Cu-Ti alloy foil patterns already measured in a high-throughput study whose acquisition parameters were optimized for lattice-parameter and phase identification. The X-ray volume-fraction distribution is converted to equivalent area-weighted fractions through a spherical-section transformation before comparison with area-weighted electron backscatter diffraction distributions. The 0.7 at.% Ti specimen, with the smallest grains and strongest texture, shows the largest discrepancy; the 4 at.% Ti and 5 at.% Ti specimens agree better, and the 5 at.% Ti specimen has the larger average grain size.

Perspective

The demonstrated scope covers synthetic diffraction patterns and experimental patterns from sputter-deposited Cu-Ti alloy foils; each reflection ring is analyzed independently, and grain size is converted from intensity to spherical-equivalent diameter through a reflection-specific volume-per-count coefficient.

For small grains, severe spot overlap prevents accurate intensity-distribution recovery, and stronger texture shifts the recovered grain-size distribution toward larger sizes; for large grains, too few spots reach the detector, so distribution recovery is limited by the number of grains in the scattering volume and can be improved by pooling multiple scans.