Prediction of Equipment Remaining Useful Life Based on Graph Learning and Spatiotemporal Knowledge Graph

ABSTRACT As equipment structures and functionalities become more complex, ensuring safety and reliability has become increasingly critical. Hence, accurately predicting the remaining useful life (RUL) of equipment has gained significant importance. Recent advances in graph learning have contributed significantly to RUL prediction by leveraging monitoring signals to extract temporal features and build predictive models. However, a key challenge persists: structured prior knowledge that describes the spatiotemporal correlations between monitoring data and equipment structure is often lacking, and relational priors are not effectively incorporated in the modeling process. To address these challenges, this paper proposes a spatiotemporal knowledge graph (STKG) modeling method for equipment, combined with a graph‐based spatiotemporal feature learning algorithm for RUL prediction. The main contributions of this work are as follows: (1) The STKG models the hierarchical relationships among equipment, sensor signals, and state transitions across both spatial and temporal dimensions; (2) A graph attention convolution‐pooling network, incorporating relational priors, is proposed to extract spatial features from the STKG at different time points, constructing spatial graph aggregation mappings; (3) The informer network is employed to capture temporal decay patterns, generating cross‐time and sensor graph representations for RUL prediction. The proposed method is validated on a public dataset, demonstrating superior performance compared to existing models.