Deep Self-Supervised Domain Adaptation Network for Fault Diagnosis of Rotating Machine With Unlabeled Data

Recently, domain adaptation (DA)-based fault diagnosis (FD) approaches have been receiving increasing attention in intelligent FD of rotating machinery due to its powerful knowledge transfer capability on different working conditions. The existing methods traditionally assume that the number of faul...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement 2022, Vol.71, p.1-9
Main Authors: Li, Jipu, Huang, Ruyi, Chen, Junbin, Xia, Jingyan, Chen, Zhuyun, Li, Weihua
Format: Article
Language:English
Subjects:
Adaptation
Clustering
Convolution
Domain adaptation (DA)
Domains
Employee welfare
Fault diagnosis
fault diagnosis (FD)
Feature extraction
Gearboxes
Handheld computers
Industrial applications
Knowledge management
Rotating machinery
Rotating machines
self-supervised
Training
Transfer learning
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, domain adaptation (DA)-based fault diagnosis (FD) approaches have been receiving increasing attention in intelligent FD of rotating machinery due to its powerful knowledge transfer capability on different working conditions. The existing methods traditionally assume that the number of fault categories between the source and target domains is equal. However, this assumption generally does not conform to the real industrial scenarios, where the coincidence degree between two domains is uncertain due to the unlabeled data obtained from the target domain. Therefore, a more general framework is studied to satisfy unclear DA scenarios. This demanding DA problem is solved in this article using deep self-supervised DA network. First, a novel neighborhood clustering technique is used to cluster fault samples in a self-supervised way. Second, an entropy-based feature alignment is introduced into the proposed method to separate the known and unknown samples in source and target domains. The experimental results on a gearbox dataset demonstrate that the proposed method is hopeful for actual industrial applications.
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2022.3164136