Designing a Digital Twin Approach Using Time–Frequency-Space Modulation Graph Fourier Transform Network for Aeroengine Health Indicators

Accurate prediction of control, status and performance parameters is crucial for the safe and reliable operation of aircraft engines. Leveraging digital technology to construct digital twin models that predict the real-world activities of physical entities can provide timely operational alerts or preventive maintenance recommendations, thereby offering technical support for prognostics health management of aircraft engines. This paper proposes a time-frequency-space modulation graph Fourier transform network (TFSGFTN) to construct digital twin models for aircraft engines. Firstly, in the time domain space, TFSGFTN utilizes a Transformer to extract temporal features by encoding the sensor time series. Subsequently, the encoded data is analyzed through Fourier transform and converted into the frequency domain. Additionally, a novel graph attention network is developed in the frequency domain space to capture the spatial features of signals. Finally, the time-frequency-space modulation model is used to construct the mirrored digital twin of the aircraft engine, which is capable of accurately predicting trends in control, status and performance indicators of the engine to ensure its healthy operation. The effectiveness of the model has been validated using publicly available datasets as well as real engine datasets from a certain institute, significantly improving prognostics health management, thereby reducing engine downtime and maintenance costs.