Multi-Feature Fusion and Reinforcement Model for High-Speed Train Axle Box Bearing Fault Diagnosis Under Variable Speed Domain.

The bogie of high-speed trains is the key component of the high-speed train. The fault diagnosis of bogie axle box bearing can ensure the safe operation of the whole vehicle. The signal obtained from the axle box bearing has the characteristics of variable speed and multiple distributions. To solve the above problems, we designed a new multi-scale feature fusion network, which is combined with a feature reinforcement mechanism. The proposed SE-MSFNet creatively combines signal processing methods with neural networks to construct different multiscale branches. Cascade convolutions with odd-even mixed dilated rates are proposed to expand the receptive field and reduce grid effects. A weight unit combined with the channel focus mechanism is designed to increase the channel feature weight conducive to classification and improve its sensitivity to fault features. The proposed method improves the feature extraction ability and robustness of the model on multiple distributed data through multi-scale feature fusion and enhancement. Compared with other advanced feature fusion methods, this method achieves better classification performance in both the full speed domain and the variable speed domain of the bogie.