Predicting the performance status of aero-engines using a spatio-temporal decoupled digital twin modeling method

Abstract

The digital twin model for predicting engine performance enhances engine health management. Key indicators such as exhaust gas temperature (EGT) and thrust are essential for evaluating engine performance. This study focuses on extracting and integrating complex spatio-temporal features from multiple sensors to construct an effective prediction model. A data-driven modeling method that combines the physical structure of an engine while achieving spatio-temporal feature decoupled was proposed. This method is based on Long Short-Term Memory (LSTM) and a self-attention mechanism, and incorporates time-variant parameter derivatives into the model’s input using first-order backward differences. Case studies were conducted on the EGT and thrust predictions. The mean absolute relative error ($MARE$) was used to evaluate the accuracy of each test, whereas the average $MARE$ (${\mu }_{MARE}$) across ten tests was used to assess the accuracy of each model. The results show that the spatio-temporal decoupled modeling method improves prediction accuracy and stability, achieving a minimum ${\mu }_{MARE}$ of 0.64% for the EGT and 0.277% for the normalized thrust. Furthermore, to test the method’s robustness against varying sampling frequencies during deployment, the sampling intervals of the test data were adjusted to simulate changes in sampling frequency. The results demonstrate that the proposed method exhibits excellent stability.