Jul 2026· Eksploatacja I Niezawodnosc-maintenance and Reliability· 0 citations· 28 references
TL;DR
Gradient-free adaptive BN gives the best risk–benefit trade-off, matching or outperforming them in five of six scenarios on C-MAPSS, and a four-step deployment workflow for cross-domain predictive maintenance is proposed.
Abstract
Deep learning models for remaining useful life (RUL) prediction degrade under varying operating conditions due to distribution shifts. Test-time adaptation (TTA) addresses this issue by updating a pretrained model using unlabeled target data without source data access; however, it has not been systematically studied for RUL prediction. Four TTA strategies are reformulated and evaluated across six cross-domain scenarios on C-MAPSS. TTA improves performance only when the source has more diverse operating conditions than the target. Restricting adaptation to batch-normalization (BN) affine parameters makes all gradient-based methods produce functionally equivalent outputs (BN affine bottleneck). Gradient-free adaptive BN gives the best risk–benefit trade-off, matching or outperforming them in five of six scenarios. A four-step deployment workflow for cross-domain predictive maintenance is proposed. The direction dependence of TTA effectiveness and the BN affine bottleneck are further confirmed on an LSTM backbone and the N-CMAPSS dataset.
Remaining useful life (RUL) estimation is a fundamental task in Prognostics and Health Management (PHM), supporting condition-based and predictive maintenance of engineering systems. Data-driven methods contribute to many effective strategies for RUL prediction. However, two problems need to be solved when they are use...
Jie-Hua Qi, Haoran Wang, Rui Wang et al.· Machine Learning and Knowled...· 0 citations
Lithium-ion batteries are widely deployed in electric vehicles and energy-storage systems, and accurate early prediction of cycle life is critical for safety, warranty cost, and second-life planning. Models trained on one dataset, however, rarely transfer to new chemistries, test protocols, or laboratories, and re-coll...
Peng-Cheng Zhao· Applied and Computational En...· 0 citations
A bearing RUL prediction method integrating online IDP detection with intelligent optimization transfer learning and a transfer prediction network combining multi-layer bidirectional long short-term memory and a convolutional residual network that captures multi-scale temporal dependencies and enhances cross-domain fea...
Tong-Shan Liu, Yi-Ming Li, Zhi-Hao Hu et al.· Measurement science and tech...· 0 citations
A novel multi-view temporal structure-aware learning framework that generates multiple perspectives of the degradation state, coupled with a Transformer-based backbone to capture long-range dependencies and a temporal ordering constraint learning mechanism to enhance stability and physical rationality.
Bo Li, Xiao-Jun Xia, Yu-Jiang Liu· Applied Sciences· 0 citations
To address heavy data reliance and poor cross-working condition generalization of rolling bearing remaining useful life (RUL) prediction models under small-sample scenarios, this paper proposes a novel method integrating meta-learning, parallel temporal modeling and the domain adversarial mechanism. Firstly, feature ex...
Xin Zhang, Jian-Fei Zheng, Hong Pei et al.· Journal of Vibration and Con...· 0 citations
D-TAIA is introduced, a framework for a joint next activity and remaining time prediction task via parameter-efficient fine-tuning of an FM backbone that combines domain-aware triplet loss pre-training with FAISS-based nearest neighbor retrieval for remaining time prediction, and adopts the TAIA inference strategy to p...
S. Van Straten, Christine Jacob, Marwan Hassani· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
