Uncovering crystal structure evolution via nanobeam X-ray diffraction with a continuity-driven machine learning approach

• Robust and generalized unsupervised continuity-driven machine learning framework for nanoXRD-based SXDM data. • Novel metric for detecting structural discontinuities without labels, usually overlooked by conventional methods. • Visualization of structural change at interfaces/boundaries in bulk GaN. • New insights linking crystal processing to microstructural defects. Nanobeam X-ray diffraction (nanoXRD) enables nanoscale mapping of crystal structures across wafer-scale crystals, offering unique insight into microstructural evolution during crystal growth. However, the resulting large and complex diffraction datasets make it challenging to quantitatively resolve local structural transitions and their connection to growth processes using conventional analysis. Here, we present a continuity-driven, unsupervised, and generalized analysis framework, referred to as the neighborhood-based similarity metric , which integrates spatial coordinates with nanoXRD data to reveal structural variations across growth sectors, interfaces, and defect-related regions without requiring prior knowledge or labels. By introducing Jaccard similarity scores to compare local neighborhoods in spatial and diffraction domains, the method quantitatively detects discontinuities where structural evolution disrupts the local continuity of diffraction patterns. Our unsupervised approach, validated with synthetic data and nanoXRD measurements of bulk GaN crystals, successfully identified both known defects and previously hidden structural discontinuities. The results provide new insights into the relationship between growth conditions, local strain evolution, and defect formation, establishing a robust and interpretable approach for linking processing and structural characteristics in complex crystalline materials.