Fault Diagnosis Method for Rolling Bearings Based on SE-MSCNN-GRU with Adaptive Sample Length Adjustment

To address the issues of uneven data collection, inaccurate fault feature extraction, and insensitivity to important bearing features in real-world scenarios, this paper proposes a fault diagnosis method for rolling bearings based on Multi-dimensional Convolutional Neural Network (MSCNN), Squeeze-and-Excitation Network (SE-Net), and Gated Recurrent Unit (GRU) with adaptive sample length adjustment. Firstly, after performing Principal Component Analysis (PCA), we selected the maximum number of principal components that achieve a 99% variance explanation rate as the final dimensionality for reduction, thus standardizing the length of the samples. Then, MSCNN extracts local features from the signal, SE-Net enhances the network's focus on key information through ‘squeeze’ and ‘excitation’ operations, and the GRU module is responsible for extracting time series features. Finally, the CWRU bearing dataset was used as the experimental subject, and the SE-MSCNN-GRU model was compared with various other models. The experimental results show that the proposed method demonstrates its excellent performance in bearing fault diagnosis, which has large application potential and advantages.