Deep Learning Enables Rapid Identification of a New Quasicrystal from Multiphase Powder Diffraction Patterns

Area of Science:

• Materials Science
• Crystallography
• Artificial Intelligence

Background:

• Over 100 stable quasicrystals have been identified since their discovery.
• Transmission electron microscopy (TEM) is the standard but labor-intensive method for quasicrystal verification.
• Rapid and automatic powder X-ray diffraction (PXRD) is needed for efficient discovery of new quasicrystals.

Purpose of the Study:

• To develop a rapid, automated technique for identifying quasicrystalline phases in powder diffraction patterns.
• To demonstrate the efficacy of deep learning (DL) for phase identification in complex multiphase samples.
• To facilitate the discovery of novel quasicrystals.

Main Methods:

• Utilized deep neural networks (DNNs) trained on artificially generated multiphase PXRD patterns.
• Applied the trained DL classifier to screen a dataset of 440 actual powder patterns.
• Focused on identifying quasicrystalline phases, even when mixed with other crystalline phases.

Main Results:

• The DL classifier achieved an accuracy greater than 92% in identifying quasicrystals from actual PXRD patterns.
• Successfully identified a new Al-Si-Ru quasicrystal from a multiphase powder sample.
• Demonstrated the capability of DL to distinguish subtle quasicrystalline signatures within complex diffraction data.

Conclusions:

• Deep learning offers a powerful and rapid tool for identifying unknown phases from PXRD data.
• This AI-driven approach significantly accelerates the search for novel quasicrystals.
• The method is effective even in challenging multiphase samples where human experts may struggle.