Local-Global Correlation Fusion-Based Graph Neural Network for Remaining Useful Life Prediction

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Remaining useful life (RUL) prediction is crucial for system prognostics and health management.
• Deep learning (DL) models excel at capturing temporal dependencies in time series data for RUL prediction.
• Existing methods struggle to model spatial dependencies from multi-sensor data, limiting RUL prediction accuracy.

Purpose of the Study:

• To propose a novel framework, LOGO (LOcal-GlObal correlation fusion), for enhanced RUL prediction.
• To effectively model and capture both local and global sensor correlations.
• To improve the accuracy of RUL prediction by addressing limitations in spatial dependency modeling.

Main Methods:

• Developed a LOcal-GlObal correlation fusion (LOGO) framework.
• Incorporated local correlations (dynamic sensor relationships) and global correlations (stable sensor relations).
• Utilized an adaptive fusion mechanism to combine correlations and defined sequential micrographs for Graph Neural Network (GNN) analysis.

Main Results:

• The LOGO framework effectively models spatial dependencies by fusing local and global sensor correlations.
• Graph neural networks capture spatial dependencies within micrographs, while temporal dependencies are captured sequentially.
• Experimental results demonstrate the effectiveness of the proposed method for accurate RUL prediction.

Conclusions:

• The LOGO framework offers a significant advancement in RUL prediction by effectively integrating local and global sensor correlation information.
• The proposed approach enhances feature learning for prognostics and health management systems.
• This method provides a more robust solution for predicting the remaining useful life of systems using multi-sensor data.