A method for predicting performance degradation trends in complex equipment based on maintenance-aware dynamic graph convolution

Accurate modeling and prediction of the aero-engine performance degradation process are important foundations for realizing condition-based maintenance and intelligent health management. Exhaust gas temperature (EGT), as an important indicator reflecting the thermal condition and gas-path performance degradation of aero-engines, is affected by various factors, including operating condition changes, environmental disturbances, sensor noise, and maintenance events under actual operating conditions. However, effectively extracting degradation-related characteristics from such complex signals remains challenging, limiting the accuracy of long-term degradation trend prediction. To address these issues, this study proposes an Evolutionary Graph Convolutional Feature Network (EGCFNet) for predicting aero-engine EGT degradation trends. Firstly, a spectral domain trend extraction module is developed to adaptively extract degradation-related frequency-domain features and suppress irrelevant disturbances, thereby enhancing degradation representation capability. Secondly, a gated decoupling temporal attention mechanism is introduced to disentangle degradation-related features from operational variations. Furthermore, a maintenance-aware dynamic graph convolutional module is developed to capture engine performance evolution by modeling dynamic correlations among multi-source monitoring parameters under maintenance-induced state transitions. Finally, a dual-branch prediction structure is designed to jointly model long-term degradation trends and local dynamic variations, thereby improving prediction accuracy. The proposed method is validated using real-world aero-engine water washing maintenance datasets and public benchmark datasets. Experimental results demonstrate that EGCFNet achieves superior prediction accuracy and robustness compared with representative baseline models, demonstrating its potential for aero-engine health monitoring and predictive maintenance applications.