Colloquia

Summary

Wind turbines are critical to renewable energy production and maintaining operational reliability is important to ensure the energy supply. This thesis investigates the potential of using SCADA data to develop a condition monitoring system for Enercon wind turbines, explicitly targeting early detection of rectifier faults in cooperation with the Enova Power GmbH. Traditional condition monitoring often requires costly additional sensors. Still, this research explores the use of existing SCADA data, processed through Long Short- Term Memory (LSTM) neural networks, to predict the rectifier fault before it occurs. The study uses accurate data from a test wind park consisting of nine Enercon E-115 EP2 wind turbines. It examines two different approaches to preprocessing SCADA data and optimizing LSTM architectures while exploring the time windows for pre-fault detection. Through this research, the ability of LSTM-based models to detect early warning signs of failures is demonstrated, achieving high levels of accuracy with an F1-score of over 98% in predicting the rectifier fault 96 hours before the actual failure. The potential impact of this approach on reducing downtime, extending turbine lifecycles, and improving the sustainability of wind turbines is also discussed. This work contributes to the growing field of predictive maintenance, especially for Enercon wind turbines in wind energy, by demonstrating the viability of using SCADA data for condition monitoring systems, offering a cost-effective solution to enhance turbine reliability for operators.