Rolling Bearing Fault Diagnosis Method Based on Deep Temporal Residual Shrinkage Network for High-Speed Trains
Year: 2025
Type: article
Abstract: Bearings in high-speed train running gears operate under high-speed, heavy-load, and high-noise conditions, making them prone to failures such as cracking, deformation, and indentation. Because these faults directly impact train operational safety, precise fault diagnosis is critical. However, existing methods face challenges in selecting parameters for feature extraction and in extracting effective features from vibration signals subject to noise interference in varying operating conditions. To address these challenges, this paper proposes a rolling bearing fault diagnosis method based on deep temporal residual shrinkage network (DTRSN). This network utilizes a temporal convolutional network (TCN) to adaptively extract spatiotemporal features from vibration signals, and incorporates a long short-term memory (LSTM) network to fuse contextual information, enabling a more comprehensive representation of fault severity in current signals. To suppress noise interference introduced by high-speed trains in special operating conditions, the DTRSN integrates a deep residual shrinkage network (DRSN) mechanism that learns adaptively thresholds to filter out noise-related features and thereby strengthens the expressive power of key features. Experiments conducted using a bearing dataset covering eight different operating conditions demonstrate that the DTRSN significantly outperforms four comparative deep learning models in diagnostic accuracy, verifying its effectiveness and robustness in bearing fault diagnosis for high-speed trains in complex noise environments.
Language: Chinese
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7125671841
