Rapid Identification of X-ray Diffraction Patterns Based on Very Limited Data by Interpretable Convolutional Neural...

Area of Science:

• Materials Science
• Artificial Intelligence
• Crystallography

Background:

• Accelerating materials discovery requires efficient analysis of large material characterization datasets.
• X-ray diffraction (XRD) is crucial for identifying materials like metal-organic frameworks (MOFs).
• Current analysis methods can be slow and labor-intensive.

Purpose of the Study:

• To develop a rapid and automatic method for identifying MOFs using XRD patterns.
• To train a convolutional neural network (CNN) model for accurate material identification.
• To demonstrate the effectiveness of data augmentation techniques in training AI models for materials science.

Main Methods:

• A CNN model was trained using a combination of theoretical and limited experimental XRD data.
• Data augmentation involved synthesizing new spectra by merging shuffled noise from experimental data with theoretical peaks.
• The model was trained on an augmented dataset of 72,864 samples and validated.
• Neighborhood component analysis (NCA) and class activation maps were used for analysis.

Main Results:

• The CNN model achieved a 96.7% identification accuracy for MOFs in the top 5 ranking on a test set.
• Data augmentation significantly increased the training dataset size for the CNN model.
• NCA confirmed that XRD samples from the same MOF cluster together.
• Class activation maps revealed the CNN's decision-making process for MOF identification.

Conclusions:

• The developed CNN model enables fast, one-to-one identification of MOFs from XRD patterns.
• Data augmentation is a powerful technique for training AI models with limited experimental data in materials science.
• This approach has potential applications for analyzing XRD patterns of various materials and data from other characterization tools.