Abstract:

To enable real-time performance monitoring and degradation assessment of aircraft engines， a digital-twin-based methodology for real-time monitoring and evaluation of engine performance degradation is proposed. A performance digital twin architecture was designed and implemented by integrating Long Short-Term Memory （LSTM） recurrent neural networks with the engine’s physical structural configuration. Baseline models of the performance digital twin were established using flight parameter data from the initial operational flights of an engine. The model demonstrates high-fidelity simulation capabilities for replicating the engine’s performance across diverse flight conditions. By feeding real-time operational parameters and flight state data into the baseline model， the real-time performance metrics of a pristine （non-degraded） engine under current operating conditions are simulated. Comparative analysis between simulated outputs and actual sensor measurements enables quantitative assessment of the engine’s instantaneous performance degradation. A case study involving 185 flight cycles validated the framework： Baseline models constructed from the first three flights achieved mean absolute relative errors below 0.98%， 0.94%， and 1.89% for rotational speed， pressure， and temperature predictions， respectively， with single-point inference time under 0.14 milliseconds， confirming the reliability of real-time digital twinning. The degradation assessment results of this method align well with traditional methods， demonstrating significant feasibility and advantages.

Key words: aero-engine, digital twin, performance monitoring, degradation assessment, LSTM, architecture-driven