Rotor-bearing system fault diagnosis method based on IFDA-VME and CWT-MHAN

A rotor bearing fault diagnosis method based on the Improved Flow Direction Algorithm (IFDA) optimized by the Variational Mode Extraction (VME) and the Continuous Wavelet Transform (CWT) - Multi Scale Hybrid Attention Network (MHAN) is proposed to address complex working environment problems, difficulty in identifying fault features due to the noise interference, and a low accuracy in fault diagnosis caused by the single information dimension of model input data and the insufficient feature extraction ability of the model. Firstly, this method uses weighted envelope spectral kurtosis as the fitness function and optimizes the center frequency and penalty factor of the VME using the IFDA. Input the optimal parameter combination into the VME to extract fault characteristic sensitive signals. Then, the one-dimension vibration signal is processed using the CWT to generate the time-frequency image data with significant fault features, which is used as the model input. And propose using the multi-scale mixed attention (MHA) for the feature extraction of data. Build a CWT-MHAN model to achieve fault diagnosis of the rotor bearing system. The experimental results show that the IFDA-VME can effectively extract fault feature-sensitive signals from signals contaminated by the strong noise, and its average feature energy ratio is significantly higher than other algorithms. The CWT-MHAN model can achieve single and composite fault diagnosis of rotor bearing systems and demonstrate a high diagnostic accuracy.