High-Speed Bearing Health Monitoring Method Based on Attention Mechanism Optimized Siamese Deep Residual Network

High-speed bearing has been widely applied in industrial machinery, and the health condition significantly affects its normal operation. However, the lack of practical condition monitoring system inevitably increases operation cost and risk simultaneously. Besides, deep learning (DL) has been regarded as a promising bearing fault diagnosis method, but available sample scale varies among factories due to signal acquisition cost, and the model performance may be hard to meet the requirements. Aiming at these problems, a novel DL framework named efficient channel attention-Siamese deep residual network-support vector machine (ECA-SDResNet-SVM) is proposed for bearing health condition recognition. Specifically, convolutional blocks are constructed and stacked in Siamese neural network (SNN) to learn features from randomly paired samples, and the ECA module is introduced to highlight sensitive components during model training; then, the SVM model is utilized to identify bearing fault status. Further comparison experimental results show that the ECA-SDResNet-SVM outperforms other compared transfer learning (TL) and DL models regardless of training sample scales or ambient noise levels, and the Acc results achieve 0.99364 ± 0.00045 at 0.45 split ratio, 0.67086 ± 0.02840 at 0.025 split ratio, and 0.70182 ± 0.01990 with additional −6-dB Gaussian white noise (GWN) at 0.3 split ratio. Further self-conducted bearing monitoring case has also validated the prominent performance of the proposed model.