DSAN: An Integrated Bearing Diagnosis Strategy With Dual-Spectral Feature Transform and Adaptive Position Correction Algorithm

This article investigates the critical role of reliable rolling bearing operation for the stability of mechanical drive systems. Addressed is a common issue in the research of rolling bearing fault diagnosis: it tends to rely on a single feature transformation technique and rudimentary data fusion strategy. The proposed dual-spectral feature transform technique overcomes the drawbacks of the existing feature transformation methods. This technique is capable of high-dimensional reconstruction of global and local features of vibration signals through multiscale feature extraction. Subsequently, a matching dual-stream adaptive network (DSAN) model was designed. The model achieves effective learning of both features through the discrete dual-channel structure. Generative adversarial training pattern and the introduction of space-to-depth feature extraction method achieve data enhancement and comprehensive analysis. This novel adaptive position correction algorithm (APCA) effectively fuses the feature information from the two channels, achieving self-correction of fault identification and optimization in the training process. Experiments show that our diagnosis strategy can effectively extract key features of the bearing operation data, thereby exhibiting significant advantages in diagnostic accuracy.