yriyazi / gearbox_fault_detection-icee2024

Gearbox Fault Detection

Gearbox Fault Detection Using Continuous Wavelet Transform and Vision Transformer (ViT)

Abstract

Managing the substantial volume of data generated by gearbox operations and the inherent complexities underscores the strategic advantage of employing deep learning methods. This study addresses fault detection in the gearbox using a Vision Transformer (ViT), whose images were prepared by continuous wavelet transformation. The dataset from the Open Energy Data Initiative (OEDI) serves as the foundation for our analysis. Data recorded by four vibration sensors, situated in the time domain and various operating loads, transform into two-dimensional images which, after some manipulations, are fed into a ViT, allowing us to tackle the classification challenge with 'Healthy' and 'Damaged' classes. Remarkably, we achieved an outstanding accuracy of 99.1554%, coupled with a reduction in the loss function to 0.040100. These results demonstrate the ViT's remarkable efficacy in detecting faults using images derived from signals recorded in the time domain.

Repository Contents

• : Contains the source code for implementing the Continuous Wavelet Transform and Vision Transformer for gearbox fault detection.

• : Contains the source code for Training the neural networks.

• /model: Contains pre-trained neural networks and results.

• : Contains the source code of the neural networks.

• : some functions and plotter.

Dependencies

• Python => 3.9
• Pytorch => 2.0
• NumPy => 1.4
• Matplotlib
• Wavelet Transform Library (PyWavelets)

Installation

1. Clone the repository:

   git clone https://github.com/yriyazi/GearBox_Fault_detection-ICEE2024.git

2. Download the dataset from here and place it in the appropriate directory.

3. Run the .\dataloader\splitter.py to generate the CWT images.

python splitter.py

4. Run the scripts to preprocess the data and train the model:

python Train.py

Usage

depending on your specific need you may need to train, fine-tune, or just use some of the pre-trained weights.

Please make sure that the config.yaml configurations machs the model_weight.pt

Using the pre-trained weights.

the pre-trained weights are in the model folder. In Test.py change the model address (directory) and feed your data to the neural network.

License

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

Citation

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

[Ali Asadian, Yassin Riyazi, and Seyyed Mossa Ayati*. 2024. " Gearbox Fault Detection Using Continuous Wavelet Transform and Vision Transformer (ViT)." ICEE, 2024]

Acknowledgments

• The Open Energy Data Initiative (OEDI) for providing the dataset.
• The contributors to the libraries and frameworks used in this project.

Feel free to reach out for any questions or collaborations!

---

This README is generated with the assistance of ChatGPT.

gearbox_fault_detection-icee2024's Issues

Something Missing?

Very interesting project. Readme.md makes reference to /code folder and some .py files that don't appear to be in the repo. Are they available?

Questions

Hello again! I'm trying to work through the code to generate the scalograms (splitter.py) but I'm having difficulty understanding what the final output should be so I can move onto the training code. There is some hard coding of the data to "Healthy" only, not "Broken" and a lot of the code is commented out.

I was hoping you could explain how you used that code to generate the training data. In other words, a procedural explanation like: "I started with these csv files and did such-and-such to them to get to a final something-or-other.

Many thanks for your time and your work.

Ron