Deep-learning framework for fully-automated recognition of TiO₂ polymorphs based on Raman spectroscopy.

Area of Science:

• Materials Science
• Computational Chemistry
• Spectroscopy

Background:

• Raman spectroscopy is crucial for mineral identification.
• Machine learning offers advanced analytical capabilities for spectral data.
• Titanium dioxide (TiO2) polymorphs have diverse industrial applications.

Purpose of the Study:

• To develop a deep learning framework for automatic identification of complex polymorph structures using Raman signatures.
• To evaluate the model's accuracy in identifying TiO2 polymorphs.

Main Methods:

• A novel framework combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks was developed.
• The model was trained and validated using the RRUFF spectral database.
• Synthesized TiO2 polymorphs were used for performance evaluation.

Main Results:

• The model achieved high confidence in identifying pure Anatase and Rutile.
• It successfully identified defect-rich Anatase and modified Rutile from their Raman spectra.
• The framework correctly identified Anatase in P25 Degussa TiO2.

Conclusions:

• The proposed deep learning framework demonstrates high accuracy for polymorph identification via Raman spectroscopy.
• This approach can significantly improve throughput and reduce costs in material analysis.
• The model's ability to identify modified and defect-rich structures shows its potential for complex material characterization.