A study on wind turbine gearbox fault diagnosis and interpretability using multisource information fusion methods

The internal structure of wind turbines is complex, and the intricate coupling effects between components during gearbox faults can introduce significant interference into the collected signals, greatly increasing the difficulty of fault diagnosis. To address the issue of low diagnostic accuracy caused by traditional methods relying on single sensors, this article proposes a multisource information fusion method for gearbox fault diagnosis in wind turbines. First, a DualStream-FuseNet (DSFN) model is designed, where the dual-branch structure enables fault feature extraction from vibration signals and stator current signals of the doubly fed induction generator. Subsequently, a weighted fusion strategy and attention mechanism are introduced to enhance the fault feature information. Experimental results show that the proposed method exhibits significant superiority in diagnostic accuracy, efficiency, and robustness. Finally, by analyzing the interpretability of the DSFN model, the article visually demonstrates the model’s ability to learn and represent fault features, revealing its feature extraction mechanism, and providing theoretical support for the application of multisource information fusion methods in industrial fault diagnosis.