Data Augmentation and Intelligent Fault Diagnosis of Planetary Gearbox Using ILoFGAN Under Extremely Limited Samples

Although the existing generative adversarial networks (GAN) have the potential for data augmentation and intelligent fault diagnosis of planetary gearbox, it remains difficult to deal with extremely limited training samples and effectively fuse the representative and diverse information. To tackle t...

Full description

Saved in:
Bibliographic Details
Published inIEEE transactions on reliability Vol. 72; no. 3; pp. 1 - 9
Main Authors Chen, Mingzhi, Shao, Haidong, Dou, Haoxuan, Li, Wei, Liu, Bin
Format Journal Article
LanguageEnglish
Published New York IEEE 01.09.2023
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects
Data augmentation
Extremely few samples
Fault diagnosis
Feature extraction
Fuses
Gearboxes
Generative adversarial networks
Generators
improved local fusion generative adversarial network (ILoFGAN)
intelligent fault diagnosis
Modules
multihead attention (MHA) mechanism
planetary gearbox
Training
Vibration analysis
Vibrations
Waveforms
Online AccessGet full text
ISSN0018-9529
1558-1721
DOI10.1109/TR.2022.3215243

Cover

More Information
Summary:Although the existing generative adversarial networks (GAN) have the potential for data augmentation and intelligent fault diagnosis of planetary gearbox, it remains difficult to deal with extremely limited training samples and effectively fuse the representative and diverse information. To tackle the abovementioned challenges, an improved local fusion generative adversarial network is proposed. Time-domain waveforms are first transformed into the time-frequency diagrams to highlight the fault characteristics. Subsequently, a local fusion module is used to fully utilize extremely limited samples and fuse the local features. Finally, a new generator embedded with multi-head attention modules is constructed to effectively improve the accuracy and flexibility of the feature fusion process. The proposed method is applied to the analysis of planetary gearbox vibration signals. The results show that the proposed method can generate a large number of samples with higher similarity and better diversity compared with the existing mainstream GANs using six training samples in each type. The generated samples are used to augment the limited dataset, prominently improving the accuracy of the fault diagnosis task.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9529
1558-1721
DOI:10.1109/TR.2022.3215243