Code for our paper "Data augmentation and intelligent fault diagnosis of planetary gearbox using ILoFGAN under extremely limited samples".
Created by Mingzhi Chen, Haidong Shao, Haoxuan Dou, Wei Li, and Bin Liu.
Paper Link: IEEE
If you find our work useful in your research, please consider citing:
@inproceedings{ILoFGAN for fault diagnosis,
title={Data augmentation and intelligent fault diagnosis of planetary gearbox using ILoFGAN under extremely limited samples},
author={Mingzhi Chen, Haidong Shao, Haoxuan Dou, Wei Li, and Bin Liu},
booktitle={IEEE Transactions on Reliability},
year={2022}
}
-
Hardware
- a GPU
-
Software
-
Ubuntu18
-
Pytorch 1.10
-
OpenCV
-
numpy
-
the original datasets can download from gear vibration dataset of University of Connecticut,planetary gearbox dataset of Southeast University. You should transform each time-domain waveform into the corresponding time-frequency spectrum. Then, the time-frequency spectrums are visualized into an RGB three-channel time-frequency diagram with pixels of 64*64 in the form of a thermal diagram.
The Processed datasets are in datasets.cwt_gearbox_jet_8_6.npy corresponds to the dataset of University of Connecticut,SW_gearbox_30_2_cwt_5_6.npy corresponds to the dataset of Southeast University.
python train.py
--conf configs/cwt_sw_gearbox_ilofgan.yaml
--output_dir results/cwt_sw_gearbox_ilofgan
--gpu 0- You may also customize the parameters in
configs. - It takes about 15 hours to train the network on a 2080Ti GPU.
python main_metric.py
--gpu 0
--dataset cwt_gearbox
--name results/cwt_sw_gearbox_ilofgan
--real_dir datasets/cwt_gearbox
--ckpt gen_00100000.pt
--fake_dir test_for_fidThe generated images will be saved in results/cwt_sw_gearbox_ilofgan/test_for_fid.
Our code is designed based on LoFGAN
The code for calculate FID is based on pytorch-fid
If you have any questions about the codes or would like to communicate about intelligent fault diagnosis, fault detection,please contact us: 1297008453@hnu.edu.cn
Mentor E-mail：hdshao@hnu.edu.cn