This repository addresses the scarcity of mechanical engineering datasets tailored for applying Machine Learning techniques in an industrial environment. The dataset provided here was not previously available on Kaggle, making it a valuable resource for the community.
The Gearbox Fault Diagnosis dataset consists of vibration data recorded using SpectraQuest's Gearbox Fault Diagnostics Simulator. The dataset captures vibrations using four sensors placed in different directions under varying loads from 0% to 90%. It encompasses two distinct scenarios:
- Healthy Condition
- Broken Tooth Condition
There are 20 files, with 10 corresponding to a healthy gearbox and ten from a gearbox with a broken tooth. Each file corresponds to a specific load, ranging from 0% to 90% in 10% increments.
This repository is organized into two main branches, each addressing a different approach to fault detection:
This branch focuses on applying Long Short-Term Memory (LSTM) networks for fault detection. LSTMs are a recurrent neural network well-suited for sequential data, making them a promising choice for analyzing time series vibration data.
This branch employs a combination of Continuous Wavelet Transform (CWT) and Convolutional Neural Networks (CNNs) for fault detection. The CWT provides a time-frequency representation of the data, which is then fed into a CNN for feature extraction and classification.
For detailed instructions on setting up and running the code in each branch, please look at the respective branch's README.md file.
If you find this work helpful or build upon it in your research, please consider citing the following paper:
[Navidreza Ghanbari, Yassin Riyazi, Farzad A. Shirazi*, and Ahmad Kalhor. 2023. "Enhanced Gearbox Fault Diagnosis with Fusion LSTM CNN Network." ISAV, 2023, Page Numbers. DOI]
This project is licensed under the terms of the MIT License. Please look at the LICENSE file for details.
Special thanks to SpectraQuest for providing the Gearbox Fault Diagnosis dataset.