Classification of crystal structure using a convolutional neural network.

Area of Science:

• Crystallography
• Materials Science
• Machine Learning
• Artificial Intelligence

Background:

• Powder X-ray diffraction (XRD) is crucial for material characterization.
• Traditional XRD analysis relies on discrete peak positions and intensities.
• Identifying crystal system, extinction group, and space group from XRD patterns can be complex.

Purpose of the Study:

• To introduce a deep machine-learning technique for classifying powder XRD patterns.
• To determine crystal system, extinction group, and space group using a convolutional neural network (CNN).
• To apply the trained CNN for symmetry identification of novel inorganic compounds.

Main Methods:

• A deep learning technique based on a convolutional neural network (CNN) was developed.
• Approximately 150,000 powder XRD patterns were used as input for the CNN.
• The CNN treated XRD patterns as images, interpreting features beyond human recognition.

Main Results:

• The CNN achieved high accuracy in classifying XRD patterns: 94.99% for crystal system, 83.83% for extinction group, and 81.14% for space group.
• The CNN architecture was automatically optimized without handcrafted feature engineering.
• The method successfully identified symmetries in unknown inorganic compounds.

Conclusions:

• Deep learning, specifically CNNs, offers a powerful alternative to traditional powder XRD analysis.
• CNNs can interpret complex patterns in XRD data, leading to accurate crystallographic classifications.
• This technique has significant potential for accelerating materials discovery and characterization.