A deep-learning technique for phase identification in multiphase inorganic compounds using synthetic XRD powder...

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Accurate phase identification and quantification are crucial for understanding complex inorganic compounds, particularly in materials discovery.
• Traditional methods for analyzing multiphase inorganic compounds can be time-consuming and challenging.
• The Sr-Li-Al-O compositional pool is a promising area for discovering new LED phosphors.

Purpose of the Study:

• To develop a rapid and accurate deep learning-based protocol for phase identification and quantification in complex multiphase inorganic compounds.
• To leverage simulated X-ray powder diffraction (XRD) data for training machine learning models.
• To validate the performance of the developed model on both simulated and experimental XRD data.

Main Methods:

• Generation of 1,785,405 synthetic XRD patterns by combinatorially mixing simulated patterns of 170 inorganic compounds within the Sr-Li-Al-O system.
• Development and training of convolutional neural network (CNN) models using the large synthetic XRD dataset.
• Testing the trained CNN model on real experimental XRD data for phase identification and quantification.

Main Results:

• The trained CNN model demonstrated prompt and accurate identification of constituent phases in complex multiphase inorganic compounds.
• Nearly 100% accuracy was achieved for phase identification when tested with real experimental XRD data.
• An accuracy of 86% was obtained for three-step-phase-fraction quantification using real experimental XRD data.

Conclusions:

• Deep learning, specifically CNNs trained on simulated XRD data, provides a powerful and efficient tool for analyzing complex multiphase inorganic compounds.
• This approach significantly simplifies intricate phase identification and quantification problems.
• The high accuracy on experimental data validates the potential of this computational method for materials science research and development, particularly for LED phosphors.