Rolling bearing fault diagnosis method based on improved gramian angular difference field and SPPF-MsCNN

Abstract

Aiming at the problem that the 2D image of the input deep learning model has lost edge information, which leads to the low accuracy of rolling bearing fault diagnosis, a fault diagnosis method based on improved gramian angular difference field (GADF) image and spatial pyramid pooling-fast (SPPF) to optimize multi-scale sliding convolutional neural network (SPPF-MsCNN) model is proposed. First, the high-frequency components of the bearing vibration signals are enhanced using the tangent sigmoid function, and GADF images are generated. Then, features are extracted using a multi-scale parallel convolution approach. Finally, introducing spatial pyramid pooling-fast avoids the feature loss issue caused by max pooling and average pooling. The experimental results demonstrate that the model achieves excellent fault recognition performance across different datasets. Furthermore, the proposed method effectively addresses the problem of low diagnostic accuracy caused by edge loss in 2D images. In addition, noise resistance experiments evaluate the robustness of the model under noisy conditions, while ablation studies confirm the rationality of the model architecture.