Skip to Content
MachinesMachines
  • Article
  • Open Access

18 November 2026

25 Pages

A Dual-Branch Spatio-Temporal Heterogeneous Graph Network with Deep Information Interaction for Remaining Useful Life Prediction of Rolling Bearings

,
,
and
Key Laboratory of Transportation Intelligent Operation and Maintenance Technology and Equipment, Ministry of Education, East China Jiaotong University, Nanchang 330013, China
*
Author to whom correspondence should be addressed.
This article belongs to the Section Machines Testing and Maintenance

Abstract

Accurate remaining useful life prediction of rolling bearings is essential for prognostics and health management of rotating machinery. However, most existing graph-based methods describe the degradation process by using homogeneous graph construction strategies, which lack an explicit learning mechanism for heterogeneous degradation information. To tackle these issues, this paper proposes a spatio-temporal heterogeneous graph network (STHGN) for RUL prediction of rolling bearings. Firstly, vibration signals are converted into correlation-guided heterogeneous graphs and feature-embedded temporal sequences to characterize the multi-source features of the degradation process. Then, the heterogeneous graph convolution branch and temporal convolutional network branch of STHGN are used to capture spatial correlations and temporal dependencies from different data representations, respectively. Eventually, a deep spatio-temporal information interaction and fusion (DSTI2F) module is built to adaptively integrate the spatial and temporal representations extracted from the two branches at each time step. The experimental results demonstrate that the proposed RUL prediction framework exhibits superior prediction accuracy and stable prediction performance across two datasets, achieving average MAE, RMSE, and R2 values of 0.050, 0.065, and 0.945, respectively, while the DSTI2F module provides qualitative visualization and interpretable evidence from both global and local perspectives.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.