Fault diagnosis of rolling bearing based on GOA-VMD-transformer model

Noise is often introduced into the vibration signal of rolling bearings during data acquisition, and directly using the raw signal for fault diagnosis can lead to low accuracy. Therefore, preprocessing of the original signal is essential. Variational Mode Decomposition (VMD) which is extensively applied to analyzing the unsteady signals. Compared to traditional Empirical Mode Decomposition (EMD), VMD effectively reduces endpoint and mode aliasing effects. But then, the capability of VMD is heavily rely on two parameters: the penalty factor and the mode number. To address this, a mean grounded on the Gazelle Optimization Algorithm (GOA) is putted forward to optimize the VMD hyperparameters, which is helpful to decrease the noise in the original signal of the bearing. The denoised signal is then used to extract shallow features, which are passed to the Transformer network for deep feature extraction. Eventually, a Support Vector Machine (SVM) is adopted for fault diagnosis. To validate the proposed Transformer-SVM model, bearing data from West Reserve University is used, and the results are compared with those from traditional SVM, CNN-SVM, and LSTM-SVM models.