Identification of Aviation Engine Performance Degradation Paths and Remaining Useful Life Prediction under Multiple Fault Modes Based on DTW-K-Medoids and Informer

The aero-engine, as the core system of an aircraft, has its performance degradation directly affecting flight safety. Accurate life prediction is crucial for its safe and stable operation. Different performance degradation modes of the engine and the trajectory differences under their impacts are one of the challenges faced in achieving universal modeling for life prediction and improving high-impact prediction accuracy. To address the above challenges, a method for adaptive timescale clustering of engine performance degradation trajectories under multiple fault modes and self-matching model life prediction is proposed. First, leveraging the advantage of Dynamic Time Warping (DTW) in measuring the similarity of curves with variable time scales, it is integrated with the K-Medoids method suitable for time-series clustering to achieve clustering of similar trajectories and performance degradation modes unaffected by trajectory length. Then, a life prediction model based on Informer is designed, utilizing its multi-head self-attention mechanism to autonomously mine sensor parameters sensitive to performance degradation and beneficial to life prediction, and to address the accuracy improvement challenge faced by medium and long-term life prediction. Verified by the C-MAPSS dataset, the proposed method has improved prediction accuracy by 25.04% compared with prediction methods without performance degradation path identification, and it has significant accuracy advantages over traditional deep learning prediction models such as LSTM and GRU.