A semi-supervised deep-learning approach for automatic crystal structure classification.

Area of Science:

• Crystallography
• Materials Science
• Artificial Intelligence

Background:

• Crystal structure solution is challenging, particularly with impurity phases, where traditional indexing methods are unstable.
• Existing deep learning methods for crystal structure identification primarily rely on labeled data, limiting their applicability.

Purpose of the Study:

• To apply a novel semi-supervised learning approach for identifying Bravais lattices and space groups of inorganic crystals.
• To leverage both labeled and unlabeled diffraction pattern data for improved crystal structure determination.

Main Methods:

• Development of a semi-supervised generative deep-learning model for crystal structure analysis.
• Training the model on both labeled (diffraction patterns with known crystal structures) and unlabeled diffraction data.
• Classification of powder diffraction patterns into all 14 Bravais lattices and 144 space groups.

Main Results:

• The semi-supervised model demonstrates superior generalization capabilities by utilizing extensive unlabeled datasets.
• The model achieves higher accuracy in classifying Bravais lattices and space groups compared to current deep learning approaches.
• Effective classification across a broader range of crystal classes than previously reported studies.

Conclusions:

• Semi-supervised learning offers a powerful strategy to overcome data limitations in crystal structure solution.
• The developed model provides a more robust and accurate method for identifying crystal structures from diffraction data.
• This approach has the potential to significantly advance materials discovery and characterization.