As industrial machinery evolves toward greater intelligence and automation, the significance of data-driven construction (DDC) continues to grow. This approach plays a crucial role in enhancing the reliability of equipment and minimizing maintenance expenses. Bearings are crucial elements in mechanical systems, and their malfunction can result in significant production delays and financial repercussions. This paper discusses the importance of bearing fault classification in data-driven maintenance and applications. It proposes an intelligent fault diagnosis method that integrates singular value decomposition (SVD), continuous wavelet transform (CWT), and a vision transformer (ViT) network. First, the SVD algorithm is applied to identify the noise components and improve the data quality. Then, the CWT technology is used to convert the denoised signal into a two-dimensional time-frequency representation (TFR) to display the faults features more intuitively. Finally, the multi-scale convolutional block attention module-ViT (MSCVIT) model is designed to perform both local and global feature extraction. The findings indicate that the mean diagnostic precision of the proposed approach on the classical bearing dataset from Case Western Reserve University is 94.2% in the migration experiment. The proposed fault classification method significantly improves the accuracy of fault detection. It can predict the potential faults of equipment in time, formulate targeted maintenance strategies, and reduce the unplanned down-time of the machine.