Adaptive Transfer Consistency Network for Multilevel Cross-Domain Fault Diagnosis of Wind Turbine Bearings

Wind turbine bearings are critical drivetrain components, and their failures can compromise operational safety, cause turbine shutdowns, and increase maintenance costs. In practical applications, wind turbine bearing fault diagnosis often involves two coexisting domain shifts: variations in rotational speed and differences between machines. Unsupervised domain adaptation (UDA) provides an effective means of transferring diagnostic knowledge from labeled source domains to unlabeled target domains. However, existing UDA studies typically address one transfer scenario at a time. Applying these methods to cross-speed and cross-machine transfer tasks remains challenging. In particular, global distribution alignment, fault-class correspondence, target class information, and prediction consistency must be coordinated. To address these challenges, we propose an Adaptive Transfer Consistency Network (ATCN). First, ATCN uses Progressive Multi-Level Adaptive Alignment (PMLA) to progressively strengthen global distribution alignment and learn a weighted combination of Euclidean and cosine distances for same-class centroid alignment, thereby reducing interdomain discrepancies while preserving fault-class correspondence. It then uses Progressive Balanced Knowledge Mining (PBKM) to delay target pseudo-label supervision during a warm-up stage and combines dynamic confidence thresholding with class-wise Top-K selection to obtain balanced target pseudo-labels for classification supervision and centroid computation. Finally, it employs Output-Level Consistency Regularization (OCR) to minimize target entropy and constrain diagnostic outputs under input perturbations. These components are jointly optimized within a unified objective. Across twelve transfer tasks from the public HIT-SM bearing dataset, using data collected on the MFS and Self test rigs at three rotational speeds, ATCN achieves mean cross-speed and cross-machine accuracies of 99.36% and 84.33%, respectively, outperforming the corresponding strongest baseline means by 8.65 and 23.38 percentage points, demonstrating its effectiveness on the investigated transfer tasks under laboratory conditions.

Authors

Institutions

Publication Details

Journal
Machines
Published
2026-11-18
DOI
https://doi.org/10.3390/machines14111367
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Adaptive Transfer Consistency Network for Multilevel Cross-Domain Fault Diagnosis of Wind Turbine Bearings

Fanghua Ye, Zhezheng Zhong
3 links
Machines
Machine Fault Diagnosis Techniques
article

Adaptive Transfer Consistency Network for Multilevel Cross-Domain Fault Diagnosis of Wind Turbine Bearings

Fanghua Ye, Zhezheng Zhong
article en

Abstract

Wind turbine bearings are critical drivetrain components, and their failures can compromise operational safety, cause turbine shutdowns, and increase maintenance costs. In practical applications, wind turbine bearing fault diagnosis often involves two coexisting domain shifts: variations in rotational speed and differences between machines. Unsupervised domain adaptation (UDA) provides an effective means of transferring diagnostic knowledge from labeled source domains to unlabeled target domains. However, existing UDA studies typically address one transfer scenario at a time. Applying these methods to cross-speed and cross-machine transfer tasks remains challenging. In particular, global distribution alignment, fault-class correspondence, target class information, and prediction consistency must be coordinated. To address these challenges, we propose an Adaptive Transfer Consistency Network (ATCN). First, ATCN uses Progressive Multi-Level Adaptive Alignment (PMLA) to progressively strengthen global distribution alignment and learn a weighted combination of Euclidean and cosine distances for same-class centroid alignment, thereby reducing interdomain discrepancies while preserving fault-class correspondence. It then uses Progressive Balanced Knowledge Mining (PBKM) to delay target pseudo-label supervision during a warm-up stage and combines dynamic confidence thresholding with class-wise Top-K selection to obtain balanced target pseudo-labels for classification supervision and centroid computation. Finally, it employs Output-Level Consistency Regularization (OCR) to minimize target entropy and constrain diagnostic outputs under input perturbations. These components are jointly optimized within a unified objective. Across twelve transfer tasks from the public HIT-SM bearing dataset, using data collected on the MFS and Self test rigs at three rotational speeds, ATCN achieves mean cross-speed and cross-machine accuracies of 99.36% and 84.33%, respectively, outperforming the corresponding strongest baseline means by 8.65 and 23.38 percentage points, demonstrating its effectiveness on the investigated transfer tasks under laboratory conditions.

MachinesVol. 14(11)
Nanchang University (CN)
Industry, innovation and infrastructure
Machine Fault Diagnosis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon