Cross-Level Fusion for Rotating Machinery Fault Diagnosis under Compound Variable Working Conditions

Official implementation of our paper: "Cross-Level Fusion for Rotating Machinery Fault Diagnosis under Compound Variable Working Conditions."

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

Paper Information

• Paper: Cross-Level fusion for rotating machinery fault diagnosis under compound variable working conditions
• Journal: Measurement
• Volume: 199
• Pages: 111455
• Year: 2022
• Publisher: Elsevier

Citation

@article{wang2022cross,
    title={Cross-Level fusion for rotating machinery fault diagnosis under compound variable working conditions},  
    author={Wang, Sihan and Wang, Dazhi and Kong, Deshan and Li, Wenhui and Wang, Huanjie and Pecht, Michael}, 
    journal={Measurement}, 
    volume={199},
    pages={111455},
    year={2022},
    publisher={Elsevier}
}

Overview

This repository provides a unified implementation of multiple few-shot learning methods for rotating machinery fault diagnosis under compound variable working conditions. The code supports various approaches including:

• Relation Networks: Traditional relation-based few-shot learning
• Prototypical Networks: Distance-based prototypical learning
• Cosine Similarity: Image-to-class metric learning
• KL Divergence: Multi-level metric learning with KL divergence
• Wasserstein Distance: Multi-level metric learning with Wasserstein distance

Features

• Unified Framework: Single codebase supporting multiple few-shot learning methods
• Multi-Domain Support: Handles both 1D time series and 2D image data
• Flexible Configuration: Easy-to-use configuration system
• Comprehensive Evaluation: Built-in testing and evaluation metrics
• Clean Architecture: Well-organized, maintainable code structure

Installation

Requirements

• Python 3.7+
• PyTorch 1.8+
• NumPy
• Matplotlib
• SciPy
• PIL (Pillow)

Setup

# Clone the repository
git clone https://github.com/your-username/multi-domain-few-shot-learning.git
cd multi-domain-few-shot-learning

# Install dependencies
pip install torch torchvision numpy matplotlib scipy pillow

Usage

Quick Start

1. Train a relation network model:

python train.py --method relation --modeltype 1d --class_num 5 --sample_num_per_class 3

2. Train a prototypical network:

python train.py --method proto --modeltype 1d --class_num 5 --sample_num_per_class 1

3. Fine-tune a pre-trained model:

python finetune.py --sample_num_per_class 10 --unfrozen 2

Run All Experiments

python run_experiments.py

Configuration

The config.py file contains all experiment configurations. You can modify parameters such as:

• Model architecture (1D/2D)
• Number of classes and samples
• Training episodes
• Learning rates
• Data types

Project Structure

multi-domain-few-shot-learning/
├── train.py                 # Main training script
├── finetune.py             # Fine-tuning script
├── run_experiments.py      # Batch experiment runner
├── config.py               # Configuration settings
├── utils/                  # Utility modules
│   ├── models.py           # Neural network models
│   ├── data_generator.py   # Data loading and preprocessing
│   └── logger.py           # Logging utilities
├── results/                # Experiment results
└── README.md               # This file

Methods Supported

1. Relation Networks

Traditional few-shot learning using relation networks to compare query and support samples.

2. Prototypical Networks

Distance-based learning using class prototypes computed from support samples.

3. Cosine Similarity

Image-to-class metric learning using cosine similarity between local descriptors.

4. KL Divergence

Multi-level metric learning incorporating KL divergence between feature distributions.

5. Wasserstein Distance

Multi-level metric learning using Wasserstein distance for distribution matching.

Data Format

The code expects PU bearing dataset with the following structure:

data/
├── KA01/
│   └── KA01/
│       ├── N09_M07_F10_KA01_1.mat
│       ├── N09_M07_F10_KA01_2.mat
│       └── ...
├── KA03/
└── ...

Results

Training results are saved in the results/ directory, including:

• Model checkpoints
• Training logs
• Accuracy plots
• Performance metrics

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

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

Acknowledgments

• Original paper authors for the research work
• PyTorch team for the deep learning framework
• Contributors to the few-shot learning community