Due to the complexity and variability of the operating environment, robust and interpretable fault diagnosis is essential to ensure the safe operation of the wind turbine gearbox. In recent years, causal learning has offered promising application prospects for uncovering the internal causal relationships of equipment and the interpretability of intelligent diagnostic models. However, the existing methods still have limitations, including difficulty in coping with the distribution offset caused by environmental changes and insufficient interpretability, which result in unreliable diagnoses. Aiming to address the above problems, an interpretable fault diagnosis method based on a causal-aware neural network (CANN) is proposed, which improves the diagnostic accuracy and interpretability of the model in complex environments. Firstly, a new structural causal model is proposed to analyze the causal relationship between fault-related variables. Then, a causal decoupling enhancement module is proposed to separate the effective causal part from the complex graph data. Finally, a robustness enhancement strategy based on causal intervention is proposed to extract stable and invariant features, which can effectively mitigate the influence of spurious correlations and distribution deviations. The experimental results show that the CANN model not only shows robust results in complex industrial environment diagnosis tasks, but also provides an interpretable explanation for model decision-making.

causal-aware,wind turbine gearbox,unknown domain,interpretable diagnosis