Automating Dislocation Characterization in 3D Dark Field X-ray Microscopy

TL;DR

This paper introduces an unsupervised method using Gram-Schmidt orthogonalization to extract dislocation features from 3D Dark Field X-ray Microscopy data, reducing dataset size while preserving critical crystallographic information.

Contribution

The work presents a novel, relatively unsupervised approach for extracting dislocation features from high-dimensional DFXM scans, improving data processing efficiency.

Findings

Reduces dataset size significantly while retaining essential information

Provides a new framework for dislocation analysis in 3D DFXM data

Enhances the interpretability of complex crystallographic scans

Abstract

Mechanical properties in crystals are strongly correlated to the arrangement of 1D line defects, termed dislocations. Recently, Dark field X-ray Microscopy (DFXM) has emerged as a new tool to image and interpret dislocations within crystals using multidimensional scans. However, the methods required to reconstruct meaningful dislocation information from high-dimensional DFXM scans are still nascent and require significant manual oversight (i.e. \textit{supervision}). In this work, we present a new relatively unsupervised method that extracts dislocation-specific information (features) from a 3D dataset ($x$, $y$, $\phi$) using Gram-Schmidt orthogonalization to represent the large dataset as an array of 3-component feature vectors for each position, corresponding to the weak-beam conditions and the strong-beam condition. This method offers key opportunities to significantly reduce dataset size while preserving only the crystallographic information that is important for data reconstruction.