Deep Neural Network Enabled Space Group Identification in EBSD.

Area of Science:

• Materials Science
• Crystallography
• Data Science

Background:

• Electron backscatter diffraction (EBSD) is crucial for materials analysis, offering multi-scale insights.
• Current EBSD indexing struggles with identifying unknown crystalline phases.
• Advancements in EBSD technology have improved data quality and collection speed.

Purpose of the Study:

• To develop a machine learning methodology for automated phase identification using EBSD patterns.
• To enable EBSD as a high-throughput technique for unknown phase identification.
• To explore the application of machine learning in classifying diffraction patterns within the (4/m, 3, 2/m) point group.

Main Methods:

• Utilized a machine learning technique for space group classification of EBSD diffraction patterns.
• Trained and evaluated the algorithm on diverse material datasets, including those outside the initial training set.
• Investigated the influence of atomic scattering factors, orientation, and pattern quality on classification accuracy.

Main Results:

• Developed a general machine learning methodology for classifying diffraction patterns by space group.
• Demonstrated the algorithm's effectiveness within the (4/m, 3, 2/m) point group.
• Evaluated the algorithm's performance in real-world scenarios, assessing the impact of various factors on accuracy.

Conclusions:

• Machine learning offers a robust approach for automated phase identification from EBSD data.
• This methodology enhances EBSD's capability for high-throughput materials analysis.
• The developed technique has the potential to replace slower, more expensive diffraction methods for phase identification.