Rapid detection of rare events from in situX-ray diffraction data using machine learning.

Area of Science:

• Materials Science
• Physics
• Engineering

Background:

• High-energy X-ray diffraction (HEXD) enables non-destructive 3D microstructure mapping of bulk polycrystalline engineering materials.
• Thermo-mechanical loading combined with HEXD captures evolving microstructures over time.
• Large data volumes and high costs of traditional methods hinder rapid analysis and temporal resolution.

Purpose of the Study:

• To present a fully automated technique for rapid detection of plasticity onset in HEXD data.
• To overcome limitations of traditional data acquisition and reduction in materials characterization.
• To enable faster extraction of actionable insights from complex experimental data.

Main Methods:

• Leverages self-supervised image representation learning and clustering.
• Transforms massive HEXD datasets into compact, semantically rich representations.
• Focuses on visually salient characteristics like diffraction peak shapes for anomaly detection.

Main Results:

• The technique is computationally over 50 times faster than traditional approaches.
• It effectively analyzes datasets that are up to nine times sparser than full datasets.
• Successfully detects anomalous events, such as changes in diffraction peak shapes, indicating plasticity onset.

Conclusions:

• The developed technique significantly accelerates the analysis of HEXD data.
• Enables just-in-time actionable information for smarter experimental design.
• Facilitates the effective deployment of multi-modal X-ray diffraction methods across various length scales.