Wind Turbine Fault Detection and Prediction Using Machine Learning Methods and SCADA Data

Condition monitoring and fault detection are critical aspects of wind farm management. With the rising number and capacity of wind farms, the need for effective wind farm maintenance strategies has gained even more importance. This paper presents an approach to detect, isolate and predict wind turbine faults using machine learning methods. Data collected from the Supervisory Control and Data Acquisition (SCADA) system, a built-in feature in most wind turbines, was used. Therefore, additional hardware costs are not required, which is a major advantage. However, SCADA systems initially were not developed for fault detection aims, resulting in imperfections in data quality. Artificial Neural Networks were used to mitigate these disadvantages and successfully interpret non-linear relations. The faults targeted in this research are non-fatal but frequent faults which reduce overall performance but do not show strong indications like catastrophic faults do. This makes their prediction from SCADA data more challenging. 5 out of 7 generator heating faults were successfully predicted as early as 56 hours in advance. The results demonstrate that the proposed approach is very powerful and significantly improve current results in this field.