Remaining Useful Life Prognostics of Bearings Based on a Novel Spatial Graph-Temporal Convolution Network

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Industrial Monitoring

Background:

• Accurate diagnosis and prediction of bearing health are crucial for modern industrial equipment.
• Data-driven prognostics show superior performance over physics-based models for remaining useful life (RUL) prediction.
• Existing methods require further advancements in accuracy and robustness.

Purpose of the Study:

• To propose a novel data-driven method for predicting the remaining useful life (RUL) of bearings.
• To leverage deep graph convolutional neural networks with spatiotemporal domain convolution for enhanced prognostics.
• To accurately identify healthy and degraded bearing states and predict RUL.

Main Methods:

• Utilized average sliding root mean square (ASRMS) as a health factor to distinguish between healthy and degraded states.
• Employed correlation coefficient analysis on hybrid features to construct a spatial graph based on feature correlation.
• Incorporated historical data for temporal convolution and processed data through spatial and temporal dimensions for RUL prediction.

Main Results:

• The proposed deep graph convolutional neural network method demonstrated high accuracy in predicting the remaining useful life of bearings.
• The spatiotemporal convolution effectively captured the complex degradation patterns of bearings.
• The graph construction based on feature correlation enhanced the model's ability to learn bearing health status.

Conclusions:

• The novel deep graph convolutional neural network with spatiotemporal convolution offers an accurate and effective approach for bearing RUL prognostics.
• This data-driven method provides a promising solution for industrial equipment health monitoring.
• The findings highlight the potential of advanced deep learning techniques in predictive maintenance.