SPINN: Physics-Informed Sparse Neural Network

Official implementation of the paper:

"Physics-Informed Sparse Neural Network for Permanent Magnet Eddy Current Device Modeling and Analysis"

D. Wang, S. Wang, D. Kong, J. Wang, W. Li and M. Pecht

IEEE Magnetics Letters, vol. 14, pp. 1-5, 2023, Art no. 2100105, doi: 10.1109/LMAG.2023.3288388.

The project has been restructured using a large model. If there are any issues with the code, please contact me.

Abstract

This paper presents a Physics-Informed Sparse Neural Network (SPINN) method for modeling and analyzing permanent magnet eddy current devices. The method combines the advantages of sparse neural networks with physics-informed constraints to solve electromagnetic field equations efficiently and accurately.

Installation

git clone https://github.com/your-username/spinn-official.git
cd spinn-official
pip install -e .

Quick Start

Official Version (Recommended)

cd spinn_official
pip install -e .
python examples/cavity_example.py

Original Version

cd code
python cavity_lubin_slip.py

Usage

from spinn_official import CavityPDE, SPINN2D, Plotter2D, App

# Create and run the application
app = App(CavityPDE, SPINN2D, Plotter2D)
app.run()

Project Structure

• spinn_official/ - Official open source code (recommended)
• code/ - Cleaned core implementation
• spinn_paper/ - Refactored version
• spinn_minimal/ - Minimal version
• code_backup/ - Original backup

Key Features

• Physics-Informed: Incorporates Maxwell equations as constraints
• Sparse Architecture: Efficient neural network design with adaptive nodes
• Multi-Region Support: Handles complex geometries with multiple materials
• Automatic Differentiation: Uses PyTorch's autograd for derivatives
• Torque Calculation: Computes electromagnetic torque from field solutions

Physical Model

The implementation solves electromagnetic field equations in a permanent magnet eddy current device with three regions:

1. Magnet Region: Permanent magnet with known magnetic field
2. Air Gap: Non-conducting air gap
3. Copper Region: Conducting copper with eddy currents

Citation

If you use this code in your research, please cite:

@article{wang2023physics,
  title={Physics-Informed Sparse Neural Network for Permanent Magnet Eddy Current Device Modeling and Analysis},
  author={Wang, D. and Wang, S. and Kong, D. and Wang, J. and Li, W. and Pecht, M.},
  journal={IEEE Magnetics Letters},
  volume={14},
  pages={1--5},
  year={2023},
  publisher={IEEE},
  doi={10.1109/LMAG.2023.3288388}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contact

For questions or issues, please open an issue on GitHub or contact the authors.