43 / 2025-03-28 18:57:55

Bearing Fault Diagnosis Based on Temporal-Spatial Convolutional Network

Temporal-Spatial Convolutional Network, detail features, overall features, multi-head attention mechanism

Deep Learning is widely used for bearing fault diagnosis due to its powerful feature extraction and learning capabilities. However, traditional deep learning algorithms do not deeply mine the dependencies and overall relationships between data before and after, thus failing to fully utilize the hidden features in the data. To address this issue, this paper proposes a novel bearing fault diagnosis model based on Temporal-Spatial Convolutional Network (TSCN). TSCN focuses on both the detailed features of data dependencies before and after, as well as the overall two-dimensional features of the data. The raw data is first normalized. After inputting the algorithm, it will enter two branch structures: temporal convolution block and spatial convolution block. The time convolution block utilizes its huge receptive field to extract the temporal dependencies of the entire data and perceive the detailed features of the data. And the spatial convolution block will map one-dimensional data to two-dimensional space through polar coordinate transformation, forming an image. Utilize the powerful image processing capabilities of Convolutional neural network to perceive the overall features of the data. Subsequently, the outputs of the two branches will undergo a multi-head-talent mechanism to enable the model to focus more on important features. The algorithm proposed can achieve an accuracy of over 99% on both datasets. In addition, compared with advanced models, our proposed TSCN is superior.