| Abstract:In order to address the issue of insufficient feature extraction and inadequate representation of bearing fault characteristics
under variable load conditions, a bearing fault diagnosis algorithm based on bilinear pooling guided feature fusion was proposed. Firstly,
preprocessing was applied to the acquired raw signal data, involving operations such as DC component removal, noise filtering, anti-
aliasing filtering, and time-domain windowing, to enhance the quality of the vibration spectrogram after signal processing. Secondly,
Fourier transformation was performed on the preprocessed signal data to compute amplitude and frequency data after transformation, fol?
lowed by the creation of corresponding vibration spectrograms. Subsequently, the Res2Net network was enhanced with channel attention
and spatial attention mechanisms to extract visual features from different points of interest, and these features were fused using a bilinear
pooling method. Finally, a classification head was constructed using fully connected layers and a softmax function to achieve bearing fault
classification. The results show that the accuracy of the proposed method in the bearing data set of Case Western Reserve University and
the Paderborn data set of Germany is 98.22% and 97.94%, respectively. In the bearing fault diagnosis, the proposed method not only
integrates the theory of automation control and the principle of control engineering in theory but also verifies its effectiveness in bearing
fault diagnosis in practice, which provides a new technical approach for early warning and intelligent diagnosis of bearing faults. |
