Wind turbine generator bearings, as core components of wind turbine systems, directly impact operational safety and power generation efficiency through their health conditions. To address the challenges of non-stationary vibration signals and strong noise interference under complex operational scenarios in wind power applications, this paper proposes a novel fault diagnosis method based on a Selective State Space Model (Vim). By incorporating global state modeling capabilities and gated selection strategies, the proposed approach overcomes the local feature limitations of traditional convolutional neural networks and the long-term dependency deficiencies of recurrent neural networks, while maintaining low computational complexity. This enables dynamic mapping between input signals and latent states with adaptive focus on critical fault features. Experimental results demonstrate that the method achieves 99.30% identification accuracy on doubly-fed wind turbine 6330 bearing data, improving accuracy by over 1.14 percentage points compared to classical models and significantly outperforming existing methods, thereby providing a high-precision solution for intelligent operation and maintenance of wind turbine systems.