Deep spiking transfer learning for rotating machinery fault diagnosis
Deep spiking transfer learning for rotating machinery fault diagnosis
Transfer learning is commonly used to find domain-invariant features between different domains. However, existing models may extract redundant and distorted features, which show poor interpretability and lead to poor performance in the target domain. To address this issue, a novel model named Multi-order Statistics Matching Gated Spiking Neural Network (MSM-HGSNN) is proposed for rotating machinery fault diagnosis. To better extract features and reduce noise in signals, an interpretable feature extraction module composed of sparse wavelet convolution and gated leaky integrate-and-fire module is proposed. Furthermore, to ensure better performance in the target domain, the multi-order statistical matching measure is proposed to align features from the source domain and target domain. Compared with traditional transfer learning methods, the proposed method demonstrates better diagnostic performance on bearing and gear datasets.
Fault diagnosis, Spiking neural network, Transfer learning, Domain adaptation, Rotating machinery.
Zhu, Peng
9db29f3d-b1f0-4277-b5f9-820e794ad160
Ma, Sai
dcf30dab-ab68-40f7-af71-7a3fa836f9d3
Han, Qinkai
6ebc06ab-c275-4f62-9d4f-e3ba8d623390
Zhu, Hongtao
f8b9286d-3ec8-4d2d-8a1a-be26a66b92fa
Xiong, Yeping
51be8714-186e-4d2f-8e03-f44c428a4a49
Zuo, Mingjian
f4e9983e-c037-49e5-b5c8-1aabdcd202b5
Chu, Fulei
bef70363-0d53-4338-a751-9af31e72dff5
Zhu, Peng
9db29f3d-b1f0-4277-b5f9-820e794ad160
Ma, Sai
dcf30dab-ab68-40f7-af71-7a3fa836f9d3
Han, Qinkai
6ebc06ab-c275-4f62-9d4f-e3ba8d623390
Zhu, Hongtao
f8b9286d-3ec8-4d2d-8a1a-be26a66b92fa
Xiong, Yeping
51be8714-186e-4d2f-8e03-f44c428a4a49
Zuo, Mingjian
f4e9983e-c037-49e5-b5c8-1aabdcd202b5
Chu, Fulei
bef70363-0d53-4338-a751-9af31e72dff5
Zhu, Peng, Ma, Sai, Han, Qinkai, Zhu, Hongtao, Xiong, Yeping, Zuo, Mingjian and Chu, Fulei
(2025)
Deep spiking transfer learning for rotating machinery fault diagnosis.
Mechanical Systems and Signal Processing, 241, [113499].
(doi:10.1016/j.ymssp.2025.113499).
Abstract
Transfer learning is commonly used to find domain-invariant features between different domains. However, existing models may extract redundant and distorted features, which show poor interpretability and lead to poor performance in the target domain. To address this issue, a novel model named Multi-order Statistics Matching Gated Spiking Neural Network (MSM-HGSNN) is proposed for rotating machinery fault diagnosis. To better extract features and reduce noise in signals, an interpretable feature extraction module composed of sparse wavelet convolution and gated leaky integrate-and-fire module is proposed. Furthermore, to ensure better performance in the target domain, the multi-order statistical matching measure is proposed to align features from the source domain and target domain. Compared with traditional transfer learning methods, the proposed method demonstrates better diagnostic performance on bearing and gear datasets.
Text
Final Accepted Manuscript_Ma&Xiong
- Accepted Manuscript
Restricted to Repository staff only until 13 October 2027.
Request a copy
Text
MSSP_ePub_13Oct2025
- Version of Record
Restricted to Repository staff only
Request a copy
More information
Accepted/In Press date: 7 October 2025
e-pub ahead of print date: 13 October 2025
Keywords:
Fault diagnosis, Spiking neural network, Transfer learning, Domain adaptation, Rotating machinery.
Identifiers
Local EPrints ID: 505824
URI: http://eprints.soton.ac.uk/id/eprint/505824
ISSN: 0888-3270
PURE UUID: 075b3769-c559-47ac-b480-5c4eccb35190
Catalogue record
Date deposited: 21 Oct 2025 16:35
Last modified: 22 Oct 2025 01:37
Export record
Altmetrics
Contributors
Author:
Peng Zhu
Author:
Sai Ma
Author:
Qinkai Han
Author:
Hongtao Zhu
Author:
Mingjian Zuo
Author:
Fulei Chu
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
