Fault Diagnosis Optimizer Benchmark

This is the repository for the benchmark study article Impact of Deep Learning Optimizers and Hyperparameter Tuning on the Performance of Bearing Fault Diagnosis.

Description

We implemented end-to-end optimization benchmark code using public bearing fault datasets and state-of-the-art fault diagnosis models. This code provides public dataset download, data preprocessing, quasi-random hyperparameter sampling, and model training.

Requirements

To use this code, we recommended to install libraries on the anaconda virtual environment. Required libraries will be installed following instructions below.

conda create -n {your virtual env name} python=3.10.6
conda activate {your virtual env name}
pip install --upgrade pip
pip install -r requirements.txt

Note: We tested this code in PC using Ubuntu Linux and CUDA GPU. Experimental specifications are listed below.

|Type|Specification| |------|---| |OS|Ubuntu 18.04| |CPU|Intel Core i9-10900K @ 3.70 GHz| |RAM|128 GB| |GPU|NVIDIA GeForce RTX 2080 SUPER x2| |CUDA version|11.2| |CUDNN version|7.6.5|

Getting Started

We provide short demo code. Check tutorial.ipynb.

License

MIT License.

Citation

If this code is helpful, please cite our paper Link:

@ARTICLE{10141610,
  author={Lee, Seongjae and Kim, Taehyoun},
  journal={IEEE Access}, 
  title={Impact of Deep Learning Optimizers and Hyperparameter Tuning on the Performance of Bearing Fault Diagnosis}, 
  year={2023},
  volume={11},
  number={},
  pages={55046-55070},
  doi={10.1109/ACCESS.2023.3281910}}