Machine Learning for Nuclear Data Validation

A comprehensive tutorial series for physics students exploring nuclear data and machine learning applications in the validation process.

📚 Overview

This project provides hands-on tutorials for downloading, visualizing, and analyzing neutron cross-section data from multiple nuclear databases. The tutorials are designed to help students understand nuclear data sources and prepare them for applying machine learning techniques to the validation process.

🎯 Learning Objectives

Understand nuclear cross-sections and their importance
Access and download data from major nuclear databases
Visualize cross-section data across different energy ranges
Compare evaluated libraries with experimental measurements
Prepare datasets for machine learning applications

📂 Project Structure

ML4ND/
├── tutorials/               # Jupyter notebook tutorials
│   ├── 00_Introduction.ipynb
│   ├── 01_ENDF_Data.ipynb
│   ├── 02_EXFOR_Data.ipynb
│   ├── 03_TENDL_Data.ipynb
│   └── 04_JANIS_API.ipynb
├── data/                   # Downloaded nuclear data (created during tutorials)
│   ├── endf/
│   ├── exfor/
│   ├── tendl/
│   └── janis/
├── requirements.txt        # Python dependencies
└── README.md              # This file

🚀 Getting Started

Prerequisites

Python 3.8 or higher
Jupyter Notebook or JupyterLab
Basic knowledge of Python and physics
~5 GB free disk space (for nuclear data files)

Installation

Clone or download this repository:
```
cd ML4ND
```

Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install required packages:
```
pip install -r requirements.txt
```
Launch Jupyter:
```
jupyter notebook
```
Navigate to the tutorials/ directory and start with 00_Introduction.ipynb

📖 Tutorial Contents

Tutorial 0: Introduction to Nuclear Cross-Sections

File: 00_Introduction.ipynb

What are nuclear cross-sections?
Types of reactions (fission, capture, elastic scattering)
Overview of nuclear data sources
Why machine learning for validation?

Duration: 20-30 minutes

Tutorial 1: ENDF (Evaluated Nuclear Data File)

File: 01_ENDF_Data.ipynb

Topics covered:

Understanding evaluated nuclear data
Downloading ENDF data from JANIS web interface
Loading and visualizing cross-section data
Understanding MT numbers (reaction types)
Comparing multiple reactions
Preparing data for machine learning

Key concepts:

Evaluated vs experimental data
MT numbers and reaction types
Energy regions (thermal, resonance, fast)
Fissile vs fissionable isotopes
Using JANIS for easy CSV export

Duration: 30-45 minutes (no installation complexity!)

Tutorial 2: EXFOR (Experimental Nuclear Reaction Data)

File: 02_EXFOR_Data.ipynb

Topics covered:

Accessing experimental measurements
Using the IAEA EXFOR database
Working with measurement uncertainties
Comparing multiple experiments
Comparing experimental data with evaluations

Key concepts:

Experimental uncertainties
Data validation
Scatter between experiments
Residual analysis

Duration: 45-60 minutes

Tutorial 3: TENDL (TALYS-based Evaluated Nuclear Data Library)

File: 03_TENDL_Data.ipynb

Topics covered:

Automated vs manual evaluations
Accessing TENDL data
Comparing TENDL with ENDF
Working with exotic isotopes
Understanding evaluation uncertainties

Key concepts:

Systematic evaluations
Extended isotope coverage
Uncertainty quantification
Evaluation methodologies

Duration: 45-60 minutes

Tutorial 4: JANIS (Multi-Library Access)

File: 04_JANIS_API.ipynb

Topics covered:

Accessing multiple libraries simultaneously
Comparing ENDF, JEFF, JENDL, TENDL, CENDL
Statistical analysis of library differences
Using library spread for uncertainty estimation

Key concepts:

Multi-library validation
International cooperation in nuclear data
Evaluation uncertainty
Ensemble methods

Duration: 45-60 minutes

🗄️ Nuclear Data Sources

🔬 Machine Learning Applications

These tutorials prepare data for various ML applications:

Anomaly Detection: Identify unusual patterns in experimental data
Interpolation: Predict cross-sections at unmeasured energies
Uncertainty Quantification: Estimate evaluation uncertainties
Data Comparison: Systematic comparison of experimental vs evaluated data
Pattern Recognition: Find systematic errors or trends
Ensemble Methods: Combine multiple evaluations

💡 Tips for Success

Start with Tutorial 0: It provides essential background
Run cells sequentially: Each notebook builds on previous cells
Explore the data: Modify parameters to test different isotopes
Save your work: Data downloads can take time
Ask questions: Nuclear data is complex - confusion is normal!

📊 Expected Outputs

Each tutorial generates:

Visualizations: High-quality plots of cross-section data
Data files: CSV files for further analysis
Comparisons: Side-by-side library and experiment comparisons

All outputs are saved to the data/ directory.

🛠️ Troubleshooting

OpenMC installation issues

# If OpenMC fails to install, try:
conda install -c conda-forge openmc

Data download issues

Some databases require registration (free)
Check your internet connection
Use university network if available (better bandwidth)

Memory issues

Reduce the number of energy points in examples
Close other applications
Consider processing data in chunks

📚 Additional Resources

Nuclear Physics

"Introduction to Nuclear Engineering" - Lamarsh & Baratta
"Nuclear Reactor Analysis" - Duderstadt & Hamilton

Nuclear Data

ENDF Format Manual: https://www.nndc.bnl.gov/endfdocs/
IAEA Nuclear Data Services: https://www-nds.iaea.org/