A Prognostic Framework for Rolling Bearing RUL Prediction Using Dynamic MAD Preprocessing and AFF‐Attention Health Indicator Fusion
Quality and Reliability Engineering International · 2026-10 · 42 (6) · 2865-2873
DOI: 10.1002/qre.70267 ↗Kurz gefasst
Fachartikel in der Zeitschrift Quality and Reliability Engineering International, online erschienen am 26.05.2026. Verfasst von Zixing Huang und 4 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Remaining useful life (RUL) prediction of rolling bearings is essential for condition‐based maintenance of complex rotating machinery. This study proposes a data‐driven prognostic framework that integrates robust signal preprocessing, adaptive feature fusion (AFF) and temporal prediction for dealing with this task. First, a systematic preprocessing pipeline is developed to improve vibration‐signal quality, including chronological concatenation, adaptive wavelet‐threshold denoising, sliding‐window segmentation, and an improved median absolute deviation (MAD) method with dynamic thresholding. By adaptively adjusting the threshold and refining the treatment of non‐numeric (NaN) segments, the improved MAD method effectively suppresses outliers, enhances the signal‐to‐noise ratio, and preserves temporal continuity. Second, a strategy combining AFF and an attention mechanism is introduced to construct a health indicator (HI) that is more sensitive to early weak degradation and more consistent with the underlying degradation trajectory. K‐means clustering is further employed to partition the health states and identify degradation stages, and piecewise linear mapping with normalization is used to generate standardized life labels. Finally, the constructed HI sequence is fed into different temporal predictors for RUL estimation. Experiments on an accelerated life‐test dataset demonstrate that the proposed AFF‐Attention‐HI‐based framework achieves higher prediction accuracy and better long‐horizon stability than conventional AFF‐HI and PCA‐HI schemes. The results further indicate that the proposed HI representation can be effectively utilized by different sequence models, providing reliable support for bearing health assessment and RUL prognostics.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Huang, Z., Sheng, W., Ren, G., Li, Y., & Huang, H. (2026). A Prognostic Framework for Rolling Bearing RUL Prediction Using Dynamic MAD Preprocessing and AFF‐Attention Health Indicator Fusion. Quality and Reliability Engineering International, 42(6), 2865-2873. https://doi.org/10.1002/qre.70267
Vancouver
Huang Z, Sheng W, Ren G, Li Y, Huang H. A Prognostic Framework for Rolling Bearing RUL Prediction Using Dynamic MAD Preprocessing and AFF‐Attention Health Indicator Fusion. Quality and Reliability Engineering International. 2026;42(6):2865-2873. doi:10.1002/qre.70267
BibTeX
@article{Huang2026,
author = {Huang, Zixing and Sheng, Wangyu and Ren, Guang and Li, Yan‐Feng and Huang, Hong‐Zhong},
title = {A Prognostic Framework for Rolling Bearing RUL Prediction Using Dynamic MAD Preprocessing and AFF‐Attention Health Indicator Fusion},
journal = {Quality and Reliability Engineering International},
year = {2026},
volume = {42},
number = {6},
pages = {2865--2873},
doi = {10.1002/qre.70267}
}
EUVIMED