Condition Monitoring of Wind Turbines and Extraction of Healthy Training Data Using An Ensemble of Advanced Statistical Anomaly Detection Models

Premature failures caused by excessive wear are responsible for a large fraction of the maintenance costs of wind turbines. Therefore, it is crucial to be able to identify the formation of these failures as early as possible. To this end, a novel condition monitoring method is proposed that uses univariate and multivariate statistical data analysis techniques to construct an anomaly detection framework based on temperature SCADA data from wind turbines. The purpose of this framework is twofold. On the one hand it should give early warnings for failures, and on the other hand it should be able to extract healthy training data from unverified data for more advanced machine learning models. A large limitation of the latter models is that they require at least one year of training data. This is necessary to avoid seasonal dependence in the sensitivity of the models. The framework developed in this research contains multiple steps. First, there is a preprocessing step in which feature engineering and data transformation happens. The second step entails anomaly detection on the temperature time series data. This method uses fleet information to filter out common factors like wind speed and environmental temperature. Multiple models are combined to get more stable and robust anomaly detections. By combining them the weaknesses of the individual models are alleviated resulting in a better overall performance. To validate the model, temperature and failure data of a real operational wind farm is used. Although the methodology is general in its scope, the validation case focusses specifically on generator bearing failures.