A temporal-spectral dual-stream anti-noise bearing fault diagnosis model based on adaptive mode decomposition.

In recent years, deep learning-based bearing fault diagnosis models have achieved remarkable performance under ideal experimental conditions. However, the accuracy and reliability of these models degrade significantly in real industrial scenarios where the acquired signals are often contaminated by strong background noise. Although a variety of anti-noise bearing fault diagnosis approaches have been developed, several limitations still remain. Most models directly take raw noisy signals as inputs, lacking an effective front-end noise suppression mechanism, which makes it difficult to sufficiently highlight fault-related information. Moreover, most models predominantly focus on the extraction and analysis of temporal features, neglecting the complementary fault characterization information embedded in other feature domains, thereby limiting the richness and discriminative power of feature representation. To address these issues, this paper proposes a temporal-spectral dual-stream anti-noise bearing fault diagnosis model based on adaptive mode decomposition. First, an adaptive mode decomposition module is designed to process raw noisy signals, aiming to suppress irrelevant noise components and enhance fault-related information, thereby providing a cleaner signal representation for subsequent diagnosis. Second, a temporal-spectral dual-stream bearing fault diagnosis framework is constructed to extract fault information from the same signal under different perspectives, aiming to enhance the richness and discriminative power of feature representation, thus the proposed model's diagnostic performance under noise interference. Finally, extensive experimental results on two real-world cases demonstrate that, compared to current mainstream anti-noise bearing fault diagnosis models, the proposed approach achieves higher diagnostic accuracy under various noise intensities, sufficiently validating its effectiveness.