SCADA-Data Analysis for Condition Monitoring of Wind Turbines

Wind energy, the world’s fastest growing renewable energy technology, is developing towards a major utility source. Turbines are growing in size and are located in more remote sites, sometimes even offshore, to benefit from better wind conditions. These developments help to maximize the output per turbine but come with challenges for operation and maintenance (O&M). Unexpected failures result in longer downtimes and consequently higher revenue losses. Hence, maintenance management promises consid-erable cost saving potential and the analysis of data form the turbine inbuilt supervisory control and data acquisition (SCADA) system can effectively support maintenance de-cisions. This thesis aims to investigate possibilities to utilize SCADA data for early failure de-tection in critical wind turbines (WTs). Therefore, a condition monitoring approach is further developed and applied. The method uses artificial neural networks to model tar-get parameters under normal operating conditions and analyzes deviations from the measured values with the help of statistical tools, such as the Mahalanobis distance (MHD) measure. In order to increase the robustness and accuracy of the approach, the development of several data pre-processing methods is presented. Two different anoma-ly detection philosophies are investigated by building two different models. A gearbox model which is monitoring local variables to indicate component malfunctions and a power model which is predicting the turbine’s power output to indicate problems form a system’s perspective. Based on the available data both monitoring approaches were applied to investigate gearbox failures for indirect drive WTs and generator bearing failures for direct drive WTs. Furthermore, the power model was found to be an effective method for ice detec-tion on WT blades. The successful detection of gearbox anomalies long before a final component breakdown is presented. However, the model was not able to detect all gear-related problems investigated. It was concluded that the availability of parameters which are potentially affected by component malfunctions play a decisive role in this approach. The power model application showed that a different anomaly detection ap-proach might be better suited for the investigated cases. However, this approach is well suited for the detection of icing and recommendations for further studies are derived