.png){kind=icon}
.png){kind=icon}
Return
CMCA: A cluster matching based class-incremental unsupervised domain adaptation method for bearing fault diagnosis
.png){kind=icon}
PRE
AI
X
Z
H
H
Y
DOI：10.1016/j.neucom.2026.134717.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In variable-condition bearing fault diagnosis, target condition data often arrive incrementally over the bearing service lifecycle. To address this, we introduce a class-incremental unsupervised domain adaptation (CI-UDA) framework that jointly aligns domains under varying label spaces while preserving knowledge across incremental time steps. Existing sample-based shared class filtering and pseudo-label assignment methods overlook the uncertainties introduced by inter-domain differences and lack a holistic, class-level decision-making approach. This leads to inaccurate shared class identification, low-precision supervisory signals, and suboptimal solutions. This paper proposes cluster matching to address these issues. Our method first calibrates the source model’s latent space via augmented source domain training, then clusters target domain features, and optimally matches these clusters to class labels using a designed matching protocol. This provides clear, precise supervision signals for the target data. Our approach is validated on three datasets, achieving an average diagnostic accuracy exceeding 95% and outperforming existing methods.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
Cited Papers