This study proposes an advanced condition monitoring and fault diagnosis method for rolling bearings of road tunnel fans, aiming to improve the safety of drivers and maintenance personnel. In view of the limitations of traditional bearing fault diagnosis techniques in terms of feature extraction and diagnostic accuracy, this paper innovatively adopts a dual-channel multiscale optimized transformer structure. With the multi-scale residual threshold module and pyramid-slicing attention (PSA) mechanism, the method can effectively extract and weight features at different scales, and then perform global correlation analysis and fault classification through the transformer network to achieve efficient diagnosis. Tests on the Case Western Reserve University bearing dataset show that the method achieves 100% accuracy on the training set and 99.74% on the validation set, which outperforms traditional methods and demonstrates excellent fault recognition capability and generalization.