Dynamic modeling-assisted tensor regression transfer learning for online remaining useful life prediction under open environment

Wentao Mao, Jiayi Wang, Ke Feng, Zhidan Zhong and Mingjian Zuo

Reliability Engineering and System Safety, 2025, vol. 263, issue C

Abstract: Dynamic online fault prognosis or prediction of the remaining useful life (RUL) of machinery with sequentially-collected monitoring data is of great significance for assurance of safety, reliability, and economic operation of engineering systems. Under open environment, however, online fault prognosis faces two challenges: (1) Distribution of degradation data tends to be inconsistent across different machines, and (2) Data distribution of the target machine may drift due to change of its operating condition. To address these two concerns, this paper takes rolling bearing as the study object and proposes a new dynamic model-assisted tensor regression transfer learning method for online RUL prediction. The key idea is to integrate the mechanism information of the physics-based simulation model and the self-supervised information of online data in the prognosis process. This proposed method includes two stages: pre-training and online prediction. In the pre-training stage, a deep tensor domain-adversarial model is constructed using offline degradation data and available online data. Meanwhile, a simulation library with different damage scales and degradation rates is established based on a five degree-of-freedom dynamic model. In the second online prediction stage, the prediction model is initialized by the pre-trained network obtained from the first stage. For each online data block collected from the target bearing, self-supervised information in terms of monotonicity is extracted through core tensor, while the data with the highest similarity is selected from the simulation library to extract mechanism information. An alternating optimization algorithm is then constructed to dynamically update the online prediction model through integrating these two kinds of information. Moreover, the paper provides a theoretical upper bound of the generalization error for model-data-fusion RUL prediction, proving that the transfer strategy utilizing mechanism information can definitely reduce the prognosis error. Experimental results on three bearing run-to-failure datasets demonstrate the effectiveness of the proposed method.

Keywords: Remaining useful life prediction; Transfer learning; Dynamic model; Self-supervised information; Generalization error upper bound (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0951832025004119
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:reensy:v:263:y:2025:i:c:s0951832025004119

DOI: 10.1016/j.ress.2025.111210

Access Statistics for this article

Reliability Engineering and System Safety is currently edited by Carlos Guedes Soares

More articles in Reliability Engineering and System Safety from Elsevier
Bibliographic data for series maintained by Catherine Liu ().