Machine Learning on Fault Diagnosis in Wind Turbines

Abstract

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