A noise-aware cross-domain degradation learning framework for remaining useful life prediction of flexible thin-walled bearings

Accurate remaining useful life prediction of flexible thin-walled bearings is critical to the reliability and safety of high-precision mechanical systems. However, reliable RUL estimation remains challenging due to scarce full-run degradation data, variable operating conditions, and strong vibration noise. Existing deep learning methods rely heavily on large-scale degradation data and lack effective cross-domain adaptation, leading to poor generalization on small-sample flexible thin-walled bearings. To address these issues, this paper proposes a noise-aware cross-domain degradation learning framework. Raw signals undergo wavelet-threshold denoising first, followed by time-frequency feature extraction. A sequential degradation module and adaptive gated Kolmogorov-Arnold network are then adopted to capture long-term degradation dependencies and nonlinear evolution patterns. A transfer learning strategy with pre-training on IEEE PHM 2012 dataset and fine-tuning with limited target data, plus a monotonicity-constrained loss function, are further introduced to embed degradation trend priors. Experiments on a self-built accelerated life test dataset show that under multiple operating conditions, the proposed method achieves an R 2 of 0.9280, root mean square error of 0.0458, mean absolute error of 0.0366, and 8.14% relative prediction error. It outperforms representative benchmarks in both accuracy and cross-domain robustness. Prediction accuracy rises steadily with more target data, with marginal gains from 60 to 100% data proportion. The framework achieves an average model inference time of 6.80 s per bearing, which lays a time foundation for local deployment in industrial bearing health management; full end-to-end implementation still needs to account for overheads from vibration data acquisition, signal preprocessing, feature extraction, RUL output and monitoring system updates.

Authors

Institutions

Publication Details

Journal
Journal of Vibration and Control
Published
2026-11-18
DOI
https://doi.org/10.1177/10775463261507990
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A noise-aware cross-domain degradation learning framework for remaining useful life prediction of flexible thin-walled bearings

Mingjun Guo, Haijiang Wu
5 links
Journal of Vibration and Control
Machine Fault Diagnosis Techniques
article

A noise-aware cross-domain degradation learning framework for remaining useful life prediction of flexible thin-walled bearings

Mingjun Guo, Haijiang Wu
article en

Abstract

Accurate remaining useful life prediction of flexible thin-walled bearings is critical to the reliability and safety of high-precision mechanical systems. However, reliable RUL estimation remains challenging due to scarce full-run degradation data, variable operating conditions, and strong vibration noise. Existing deep learning methods rely heavily on large-scale degradation data and lack effective cross-domain adaptation, leading to poor generalization on small-sample flexible thin-walled bearings. To address these issues, this paper proposes a noise-aware cross-domain degradation learning framework. Raw signals undergo wavelet-threshold denoising first, followed by time-frequency feature extraction. A sequential degradation module and adaptive gated Kolmogorov-Arnold network are then adopted to capture long-term degradation dependencies and nonlinear evolution patterns. A transfer learning strategy with pre-training on IEEE PHM 2012 dataset and fine-tuning with limited target data, plus a monotonicity-constrained loss function, are further introduced to embed degradation trend priors. Experiments on a self-built accelerated life test dataset show that under multiple operating conditions, the proposed method achieves an R 2 of 0.9280, root mean square error of 0.0458, mean absolute error of 0.0366, and 8.14% relative prediction error. It outperforms representative benchmarks in both accuracy and cross-domain robustness. Prediction accuracy rises steadily with more target data, with marginal gains from 60 to 100% data proportion. The framework achieves an average model inference time of 6.80 s per bearing, which lays a time foundation for local deployment in industrial bearing health management; full end-to-end implementation still needs to account for overheads from vibration data acquisition, signal preprocessing, feature extraction, RUL output and monitoring system updates.

Journal of Vibration and Control
Guangxi University of Science and Technology (CN)
Natural Science Foundation of Guangxi Zhuang Autonomous Region, Innovative Research Group Project of the National Natural Science Foundation of China, Middle-aged and Young Teachers' Basic Ability Promotion Project of Guangxi
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