Abstract: To prevent early weak faults in rolling bearings from evolving into severe faults, this paper proposes a new fault detection method. Initially, principal component analysis（PCA） is employed for feature selection of vibration signals to reduce data dimensions, simplify the structure of vibration data, and enhance feature expressiveness. Then, the complete ensemble empirical mode decomposition with adaptive noise algorithm（CEEMDAN） is used to decompose weak fault vibration signals interfered by background noise. By introducing adaptive noise on top of empirical mode decomposition（EMD）, this method enhances the recognition of weak features, separating trend and noise data to improve the accuracy of fault diagnosis. Finally, the transformer model is incorporated to further optimize feature extraction and representation, achieving efficient processing of long sequence data for weak fault feature extraction and characterization. This comprehensive approach, with its advantages in dimension reduction, noise suppression, and long sequence processing, holds promise for significant achievements in fault detection of rolling bearings.