Multiscale Fault Prediction and Diagnosis System for Aero-Engine Combustor Based on the Fusion of Digital Twin and Machine Learning

Faults in aero-engine combustors involve the coupling of multiphysics fields across multiple scales, highlighting the necessity of multiscale modeling and analysis. The traditional diagnosis method based on a physical model may fail to fully characterize the dynamic characteristics, resulting in a high false positive rate. This paper constructs a fault prediction and diagnosis system that cooperates with a digital twin and spatial-temporal graph convolutional network (ST-GCN) to achieve high-precision real-time diagnosis. A three-dimensional digital twin model integrating combustion flow field, temperature field, and pressure field is constructed, and dynamic multiscale feature extraction is performed in combination with an ST-GCN. The Kalman filter is used to realize the coordinated optimization of the physical model and the data-driven model. Experiments show that the system has a false negative rate (FNR) of 0.159 for diagnosing uneven airflow distribution faults; the diagnosis delay for six types of faults is 8.02–11.26 ms; the prediction accuracy is above 93.22% in a high-noise environment with a signal-to-noise ratio (SNR) of 10 dB. The system effectively solves the problem of precise diagnosis of multiphysics field coupling faults and provides a new technical architecture for aero-engine management.