Predictive maintenance of abnormal wind turbine events by using machine learning based on condition monitoring for anomaly detection

The predictive maintenance of wind turbines has become a critical issue with the rapid development of wind power generation. The early detection of abnormal operation conditions can prevent failure status, which takes a long time to recover. Energy waste can also be reduced while maintenance efficiency can be improved by using a supervisory control and data acquisition (SCADA) system to monitor the operation status of wind turbines. Massive data are generated from different sensors during wind turbine operation, and SCADA can be used to gather reports about hundreds of possible abnormal conditions. The popular maintenance methods have been mostly designed on the basis of statistical analysis and data mining. However, such schemes need not only big data but also sophisticated processing techniques. This study addresses the aforementioned challenges by proposing a deep learning model with comprehensive data preprocessing and hyperparameter tuning on batch size to achieve abnormal early detection. The necessary data preprocessing is initially conducted besides the conventional data cleaning and normalization steps, and time-series data windowing and label settings are also performed. Then, the imbalanced classes in the records are addressed by adopting an augmentation scheme called the synthetic minority oversampling technique. Principal component analysis is also used to enhance the training. Finally, the proposed deep learning method with fine-tuning is compared with three machine learning models for early anomaly event detection. Experimental results show that the proposed scheme can identify potential faults 72 hours before they occur, and the precision rate exceeds 90 %.