A vibration transfer path analysis based TFR-ViT model for fault diagnosis of aero-engine rotor

In the rotor condition monitoring of aero-engines, the presence of the casing dictates that vibration measurement locations are typically distant from the actual fault source. Given the complex transmission paths, rotor fault diagnosis is significantly hindered by signal attenuation, multi-source interference, and the scarcity of labelled fault samples. To address these challenges, a rotor fault diagnosis method based on Time-Frequency Representation (TFR) and Vision Transformer (ViT) is proposed. In this framework, Short-Time Fourier Transform (STFT) is applied to represent raw signals in the time-frequency domain. The ViT model is specifically employed to overcome transmission path interference. By utilising its inherent Self-Attention mechanism, global dependencies are captured and higher diagnostic weights are dynamically assigned to fault-related features while suppressing background noise. To mitigate training costs and the limitations of small datasets, pre-trained models are introduced, and the impact of varying network depths on diagnostic accuracy is systematically evaluated. The proposed method is validated through simulation studies on a gas generator rotor system, and an aero-engine simulation models under diverse speed conditions. Finally, the experimental validation of unbalance fault diagnosis on a real aero-engine rotor demonstrates that even under complex transmission paths, the proposed method achieves high-accuracy diagnostic performance.