Wide-beam Multi-Grain Indexing of Laue Diffraction for Neutron Tomography Experiments

by Dr Tomasz Kacprzak (PSI)

Wednesday 10 May 2023, 11:30 → 13:00 Europe/Zurich

https://psich.zoom.us/j/63940901584?pwd=OTVuS3RSY3QyS0ZZakcrZU4rQ2EyQT09 Meeting-ID: 639 4090 1584 Kenncode: 540903 (OHSA/E13)

Abstract: 
New experimental techniques enable studying complex multi-grain polycrystalline samples. The use of wide beams can greatly speed up the experimental time. However, the processing of Laue spot patterns from such experiments is challenging, as the patterns from hundreds of grains overlap and are difficult to disentangle in the absence of information about the wavelength that gives rise to the Bragg’s peak. In effect, the indexing for this problem has three components: finding the number of grains (similar to a clustering problem), finding the positions and orientations of grains (optimization problem) and spot-to-grain matching (assignment problem). We propose a novel method to tackle these problems within a single framework based on convex optimization and computational optimal transport. We model the likelihood of the observed spots and find its maximum using coordinate descent with closed-form expressions for steps in all variables. This enables indexing of samples with ~1000 grains effectively, even for noisy spot measurements and large fraction of outliers. The algorithm utilizes parallel processing on modern high-memory GPUs, which enables rapid indexing, with time-to-solution in around an hour.