Machine Learning on Fault Diagnosis in Wind Turbines

With the improvement in wind turbine (WT) operation and maintenance (O&M) technologies rise of O&M cost, fault diagnostics WTs based on a supervisory control data acquisition (SCADA) system has become among cheapest easiest methods to detect faults WTs.Hence, it is necessary monitor change real-time parameters from WT action could be taken advance before any major failures. Therefore, SCADA-driven diagnosis machine learning algorithms been proposed this study by comparing performance three different algorithms, namely k-nearest neighbors (kNN) with bagging regressor, extreme gradient boosting (XGBoost) an artificial neural network (ANN) condition monitoring gearbox oil sump temperature. Further, also compared two feature selection methods, Pearson correlation coefficient (PCC) principal component analysis (PCA), hyperparameter optimization optimizing models, grid search, random search Bayesian optimization. A total 3 years SCADA located France have used verify selected method. The results showed kNN PCA provides best R2 score, lowest root mean square error (RMSE). trained model can potential at least 4 weeks advance. However, Support Vector Machine hybrid algorithm improve its reduce alarm.