A multi-scale dynamic convolution ResNet intelligent diagnosis method based on STFT
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Fault diagnosis of rolling bearings is critical for the stable and reliable operation of rotating machinery. With advances in deep learning technology, data-driven intelligent fault diagnosis has gained increasing attention. However, constrained by the diversity of actual operating environments, existing methods still require improvement in diagnostic accuracy, generalization capability, and network stability. To address these challenges, this paper proposes an intelligent fault diagnosis method based on Short-Time Fourier Transform (STFT) and Multi-Scale Dynamic Convolution within the ResNet (MSODResNet) framework. First, STFT is employed to transform the segmented vibration signals into time-frequency images. Considering the limited feature extraction capability of the standard ResNet18 residual modules, three key improvements are introduced: First, a feature extraction strategy based on multi-scale dynamic convolution is adopted to capture fault features at different scales. Subsequently, intra-residual block cross-branch connections are implemented to preserve subtle fault features and prevent information loss. Finally, the enhanced residual network architecture integrates a cross-attention mechanism with network configuration. Testing on the CRWU bearing dataset and Jiangnan University bearing dataset achieved average diagnostic accuracies of 99.56% and 90.22%, respectively. Compared to other networks, this approach demonstrates significant advantages in diagnostic precision, stability, and noise resistance.
Fields of Study & Subjects
Machine Fault Diagnosis Techniques
Fault Detection and Control Systems
Gear and Bearing Dynamics Analysis
Publication details
- Journal / source
- Engineering Research Express
- Volume
- 8
- Issue
- 7
- Pages
- 075523–075523
- Publication date
- 1 April 2026
- Language
- EN