Spatio-Temporal Information Fusion with Graph Neural Networks for Aero-Engine Remaining Useful Life Prediction

Aircraft engines, as highly complex thermodynamic composite power systems, are subject to extreme operating conditions that render them vulnerable to failures, such as fatigue. Predicting the Remaining Useful Life (RUL) of these engines allows for the implementation of more precise predictive maintenance strategies, which can significantly reduce the risk of unexpected failures and the potential losses they may incur. Most data-driven methods primarily rely on extracting temporal feature information to construct mappings, often neglecting the inter-sensory correlation within the data samples. To address this, we introduce a graph neural network combined with attention mechanism for predicting RUL of aircraft engines. By modeling data samples as spatiotemporal graph data and integrating this information, our method fully exploits the historical degradation data of aircraft engines for predictive purposes. Ultimately, by conducting experiments on the C-MAPSS dataset and comparing our method with other data-driven approaches, we have demonstrated that our proposed method can more accurately and effectively fulfill the task of lifespan prediction.