Crystallographic phase identifier of a convolutional self-attention neural network (CPICANN) on powder diffraction...

Area of Science:

• Materials Science
• Crystallography
• Artificial Intelligence

Background:

• Spectroscopic and diffraction data are crucial for materials characterization, providing detailed crystallographic information.
• Current methods for crystallographic phase identification are often time-consuming, hindering rapid analysis.

Purpose of the Study:

• To develop a real-time crystallographic phase identification tool to overcome the limitations of current time-consuming methods.
• To enhance the speed and accuracy of materials characterization using diffraction data.

Main Methods:

• Development of a convolutional self-attention neural network (CPICANN) for real-time crystallographic phase identification.
• Training the model on a large dataset of 692,190 simulated powder X-ray diffraction (XRD) patterns from 23,073 unique inorganic crystallographic information files.

Main Results:

• CPICANN achieved high accuracies in single-phase identification on simulated XRD patterns: 98.5% with elemental information and 87.5% without.
• Bi-phase identification accuracies reached 84.2% (with elemental info) and 51.5% (without) on simulated data.
• In experimental settings, CPICANN demonstrated 80% accuracy, outperforming JADE software's 61% accuracy.

Conclusions:

• CPICANN offers superior performance in crystallographic phase identification compared to traditional software like JADE.
• Integration of CPICANN into XRD refinement software promises to significantly advance materials characterization technology.