Bearing Remaining Useful Life Prediction based on TSDAE and Pathformer-TCLSTM

The remaining useful life (RUL) prediction of bearings is crucial for the stable operation and effective maintenance. Conventional RUL prediction approaches extract the restricted features that would affect the prediction results, and the computational efficacy is often influenced by the redundancy...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:International journal of prognostics and health management Jg. 16; H. 2
Hauptverfasser: Fu, Guanghua, Yang, Yujie, Liu, Yonghui, Wang, Xuegen
Format: Journal Article
Sprache:Englisch
Veröffentlicht: 06.10.2025
ISSN:2153-2648, 2153-2648
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The remaining useful life (RUL) prediction of bearings is crucial for the stable operation and effective maintenance. Conventional RUL prediction approaches extract the restricted features that would affect the prediction results, and the computational efficacy is often influenced by the redundancy of the features and domain knowledge. To address these problems, this paper proposes a RUL prediction approach mainly based on temporal sparse denoising autoencoders (TSDAE) for feature selecting, and Pathformer-temporal convolutional long short-term memory (Pathformer-TCLSTM) for predicting. Firstly, the original signal is denoised via wavelet thresholding. Subsequently, the denoised signal is decomposed using empirical mode decomposition (EMD) to extract the features of the time-domain, frequency-domain, and time-frequency domain to resolve the problem of restricted features. Moreover, the TSDAE feature selection technique is implemented to eliminate redundant features and address the limitation of domain knowledge utilized in traditional feature selection. Finally, the Pathformer-TCLSTM model is adopted for RUL prediction, which captures the multi-scale global information, local information, and long-range dependency. The validation on the PHM2012 and XJTU-SY bearing datasets shows that the proposed model has satisfactory predictive performance.
ISSN:2153-2648
2153-2648
DOI:10.36001/ijphm.2025.v16i2.4297