Insightful classification of crystal structures using deep learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Crystallography

Background:

• Automated knowledge extraction is vital for data-driven materials science.
• Accurate crystal symmetry identification is essential for materials characterization.
• Existing methods struggle with defective structures and require user-defined thresholds.

Purpose of the Study:

• To develop a machine learning-based approach for automatic crystal structure classification by symmetry.
• To overcome limitations of current methods in handling defective and noisy materials data.

Main Methods:

• Representing crystal structures using calculated diffraction images.
• Constructing and applying a deep learning neural network model for classification.
• Utilizing attentive response maps to interpret the neural network's decision-making process.

Main Results:

• Successfully classified over 100,000 simulated crystal structures, including heavily defective ones.
• The deep learning model identified relevant structural features analogous to human expert analysis.
• Demonstrated the model's ability to generalize and interpret complex crystallographic data.

Conclusions:

• The proposed machine learning approach enables robust and automated crystal symmetry recognition.
• This method is suitable for analyzing large datasets of potentially noisy or incomplete 3D structural data in materials science.
• Paves the way for advanced big-data analytics in materials discovery and characterization.