NeuralXRD

This is the official code for ICLR'23 ML4Materials Workshop paper "Machine Learning-Assisted Close-Set X-ray Diffraction Phase Identification of Transition Metals".

Overview

• xrd_patterns: a folder containing a batch of XRD diffraction patterns in tsv format. All of the phases come from Crystallography Open Database and were extracted using open Mercury software.
• data_generation.py: functions to create a dataset of synthetic phases.
• integration_algorithm: main functionality of the proposed method. File includes functions to compute phase stats and match elements.
• calibration.py: training of calibration models. They help in making corrections to the integration method results.
• utils.py: utility functions.
• demo.py: an example of the running code.

How to run the code

In order to run an experiment, you need to create your DataFrame using functions from data_generation.py and then apply integration algorithm. Simple example (demo.py):

from data_generation import create_df_from_xrd_files, generate_synthetic_phases
from integration_algorithm import score_method
from utils import normalize_intensity, intensities_to_list


if __name__ == '__main__':
    # Acquire the data
    data = create_df_from_xrd_files(path_to_xrd_files='xrd_patterns')
    data = generate_synthetic_phases(data)

    # Preprocess 
    data = normalize_intensity(data)
    data = intensities_to_list(data)

    # Evaluate
    score_method(data, save_experiments=False, calibration_model=None)

Reference

To cite this paper use the following reference:

@inproceedings{zhdanov2023machine,
  title={Machine learning-assisted close-set X-ray diffraction phase identification of transition metals},
  author={Zhdanov, Maksim and Zhdanov, Andrey},
  booktitle={Workshop on''Machine Learning for Materials''ICLR 2023}
}