Multi-expert Attention Network with Unsupervised Aggregation for long-tailed fault diagnosis under speed variation

作者：

Highlights：

• A Multi-expert Attention Network with Unsupervised Aggregation (UA-MAN) was proposed for long-tailed fault diagnosis under speed variation.

• Multi-expert network is designed to learn capabilities of tackling different class distributions from the single long-tailed train dataset.

• Swin transformer block is adopted as the backbone for each expert network to suppress domain shift caused by speed variation.

• The performance of UA-MAN is verified with two comparative case studies under speed variation with different imbalanced distributions.

摘要

•A Multi-expert Attention Network with Unsupervised Aggregation (UA-MAN) was proposed for long-tailed fault diagnosis under speed variation.•Multi-expert network is designed to learn capabilities of tackling different class distributions from the single long-tailed train dataset.•Swin transformer block is adopted as the backbone for each expert network to suppress domain shift caused by speed variation.•The performance of UA-MAN is verified with two comparative case studies under speed variation with different imbalanced distributions.

论文关键词：Fault diagnosis,Speed variation,Long-tailed distribution,Unsupervised learning

论文评审过程：Received 21 April 2022, Revised 30 June 2022, Accepted 6 July 2022, Available online 10 July 2022, Version of Record 19 July 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109393