████████████╗╗    ██████████╗╗  ████████████╗╗  ████╗╗████████████╗╗
████████████╗╗    ██████████╗╗  ████████████╗╗  ████╗╗████████████╗╗
  ████╔╔════████╗╗████╔╔════████╗╗████╔╔════████╗╗████║║████╔╔════████╗╗
  ████╔╔════████╗╗████╔╔════████╗╗████╔╔════████╗╗████║║████╔╔════████╗╗
  ████████████╔╔╝╝██████████████║║████████████╔╔╝╝████║║████║║    ████║║
  ████████████╔╔╝╝██████████████║║████████████╔╔╝╝████║║████║║    ████║║
  ████╔╔════████╗╗████╔╔════████║║████╔╔══════╝╝  ████║║████║║    ████║║
  ████╔╔════████╗╗████╔╔════████║║████╔╔══════╝╝  ████║║████║║    ████║║
  ████║║    ████║║████║║    ████║║████║║          ████║║████████████╔╔╝╝
  ████║║    ████║║████║║    ████║║████║║          ████║║████████████╔╔╝╝
╚╚══╝╝    ╚╚══╝╝╚╚══╝╝    ╚╚══╝╝╚╚══╝╝          ╚╚══╝╝╚╚══════════╝╝

RAPID — Rietveld Analysis Pipeline with Intelligent Deep-learning

RAPID automates Rietveld refinement of powder X-ray diffraction (XRD) data using convolutional neural networks. A CNN predicts crystallographic and profile parameters from a diffraction pattern in a single forward pass (~2 ms), then initializes automated FullProf refinement. The three-stage pipeline covers data augmentation, CNN training, and automated refinement.

Please cite the following paper if you use this code in your research:

@article{Mun:yr5164,
    author = "Mun, Suk Jin and Nam, Yoonsoo and Choi, Sungkyun",
    title = "{Automation of Rietveld refinement through machine learning}",
    journal = "Journal of Applied Crystallography",
    year = "2026",
    volume = "59",
    number = "2",
    pages = "",
    month = "Apr",
    doi = {10.1107/S1600576726001494},
    url = {https://doi.org/10.1107/S1600576726001494},
}

Requirements

This code requires Python 2.7 (for AutoFP/FullProf refinement), Python 3.11 with PyTorch, NumPy, SciPy, and matplotlib (for CNN training and inference), and the FullProf Suite. Windows only.

Installation

To install via conda, create the two environments using the provided environment files.

Python 2.7 environment (for AutoFP data augmentation and Rietveld refinement):

conda env create -f environment_py27.yml
conda activate rapid_py27

Python 3.11 environment (for CNN training and inference):

conda env create -f environment_py311.yml
conda activate rapid_py311

For detailed setup instructions, see INSTALLATION_SETUP_GUIDE.md.

Usage

1. Data Augmentation: Place .cif and .dat files in dat_vestacif_files/, configure inputs.txt, run data_augmentation.bat
2. CNN Training: Configure CNN/macro_inputs/ML_inputs_1.txt, run CNN/train_CNN_macro.bat
3. XRD Analysis: Run xrd_analysis.bat to identify unknown materials and perform automated refinement

For detailed usage instructions, input file formats, and configuration options, refer to RAPID_manual.pdf.

Acknowledgments

This project uses AutoFP for automated Rietveld refinement and the FullProf Suite as the refinement engine.