Loading…

An online transfer learning-based remaining useful life prediction method of ball bearings

In recent years, many artificial intelligence-based approaches are proposed for remaining useful life (RUL) prediction of bearings. However, most existing studies neglected the following problems: (1) Run-to-failure data of bearings of are generally less available; (2) Degradation trends of bearings...

Full description

Saved in:
Bibliographic Details
Published in:Measurement : journal of the International Measurement Confederation 2021-05, Vol.176, p.109201, Article 109201
Main Authors: Zeng, Fuchuan, Li, Yiming, Jiang, Yuhang, Song, Guiqiu
Format: Article
Language:English
Subjects:
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:In recent years, many artificial intelligence-based approaches are proposed for remaining useful life (RUL) prediction of bearings. However, most existing studies neglected the following problems: (1) Run-to-failure data of bearings of are generally less available; (2) Degradation trends of bearings under different working conditions are diverse; (3) Unlabeled data of bearings acquired in the online stage have not been taken into account. To solve these problems mentioned above, an online transfer learning method is proposed. In the offline stage, a deep learning model is established through semi-supervised training to align feature spaces of representations from different domains. Then, in the online stage, unlabeled data from target domain are utilized to fine-tune parameters of the established model. Finally, RUL of specified bearings can be estimated precisely by the established model. The effectiveness and superiority of the proposed method in transfer prognostics tasks of bearings are verified by case studies. •A novel method is proposed for transfer prognostics tasks.•Data from different domains are utilized in semi-supervised learning.•Parameters of model are further fine-tuned in the online phase.•Superiority of the proposed method is verified.
ISSN:0263-2241
1873-412X
DOI:10.1016/j.measurement.2021.109201