A fault diagnosis method for wind turbine rolling bearings is proposed, which combines the gradient dynamic selective alignment （DSA） method, dual-channel feature fusion technique （DCFF）, and deep learning. Firstly, the acquired one-dimensional time series data are transformed into two different two-dimensional image representations using recursive mapping （RP） and Markov transfer field （MTF）. These transformed images are then input into an encoder for reconstruction into a low-dimensional representation. Next, a dual-channel feature fusion convolutional neural network （DCFFC） is constructed, embedding an improved attention mechanism within FRNet to extract in-depth feature information. A second-level feature fusion is performed to complete the feature classification. Finally, a gradient masking matrix is constructed using the data reconstruction loss of the decoder and the feature extraction network classification loss to achieve gradient dynamic self-adaptation throughout the fault diagnosis process. Experimental validation using the Paderborn dataset demonstrates that the proposed method achieves a diagnostic accuracy of 99%, effectively extracting fault-specific diagnostic information.

fault diagnosis  /  rolling bearing  /  wind power generator  /  convolutional neural network  /  gradient adaptive