Data-Enhanced Component Degradation Digital Twin for Aero-Engine Gas-Path Parameter Prediction

Component degradation in aero-engines can cause local model mismatch. Insufficient coordination between component-response updates and engine-level prediction correction may limit improvements in gas-path prediction accuracy. Therefore, a data-enhanced component degradation digital-twin framework is proposed for aero-engine gas-path parameter prediction. It connects degraded component-response updates with engine-level prediction correction through selective component replacement, gas-path and shaft interface coupling, and engine-level prediction coordination. Under prescribed low-pressure compressor (LPC) efficiency degradation, a multi-self-attention long short-term memory (MA-LSTM) network replaces the LPC input–output mapping. The physics-based representations of the remaining components are retained, forming a data-driven model integrated-mechanism model (DMIM). The learned component response is connected to downstream gas-path modules and shaft calculations through mass flow rate, total temperature, total pressure, total specific enthalpy and output torque. DMIM predictions are then coordinated with predictions from a parallel engine-level data-driven model using an altered Kalman filter strategy. The framework is evaluated using six simulated LPC-efficiency degradation scenarios comprising two degradation magnitudes and three degradation durations. The proposed framework achieves a mean coefficient of determination (R2) of 0.94, a mean absolute percentage error (MAPE) of 2.3%, and a maximum MAPE across predicted parameters of 3.9%. Compared with the original physics-based model and the standalone MA-LSTM model, the mean MAPE decreases by approximately 61.0% and 36.1%, respectively. These results indicate that coordinated component-level updates and engine-level prediction can improve overall gas-path parameter prediction under the specified degradation scenarios.