Application of shapley-value based explainable AI in health monitoring and fault localization for wind turbine gearboxes

In the field of wind power generation, the health status of wind turbine gearboxes directly impacts the power output of wind turbine units. Current gearbox fault diagnosis and localization techniques, which are based on domain knowledge and data-driven approaches, are constrained by the completeness of domain knowledge, insufficient data volume, and lack of algorithm transparency. To address this issue, we propose an explainable AI framework that possesses both learning capabilities and provides interpretable outputs. By incorporating the Shapley value analysis method into unsupervised and supervised learning algorithms, the framework achieves improvements, alleviating the model’s excessive dependence on data volume and enhancing the model’s interpretability. The effectiveness of the proposed framework was validated through experiments on two typical wind turbine gearbox cases. The results of case 1 indicate that, compared to unsupervised and supervised learning algorithms, the proposed framework significantly improves clustering performance in situations with scarce data labels. The results of case 2 demonstrate that the framework, through model interpretability analysis, achieves the localization of wind turbine gearbox fault causes, providing guiding suggestions for gearbox fault prevention and maintenance. The experimental results showcase the significant effectiveness of the ‘knowledge + data’ integration approach in engineering applications, offering valuable references for the practical implementation of explainable artificial intelligence.