Cost-Sensitive LightGBM-Based Online Fault Detection Method for Wind Turbine Gearboxes

In practice, faulty samples of wind turbine (WT) gearboxes are far smaller than normal during operation, and most the existing fault diagnosis methods for WT only focus on improvement classification accuracy ignore decrease missed alarms reduction average cost. To this end, a new framework is proposed through combining Spearman rank correlation feature extraction cost-sensitive LightGBM algorithm gearbox’s detection. article, features from supervisory control data acquisition (SCADA) systems firstly extracted. Then, selection employed by using expert experience coefficient to analyze between big gearboxes. Moreover, detection established optimizing misclassification The false alarm rate gearbox under different working conditions finally obtained. Experiments have verified that method can significantly improve accuracy. Meanwhile, consistently outperform traditional classifiers such as AdaCost, GBDT, XGBoost in terms low rate. Owing its high Matthews scores cost, (CS LightGBM) preferred imbalanced practice.