In the pursuit of cleaner energy, wind power has become a cornerstone of the modern energy landscape. Yet, ensuring the reliability and maintainability of wind farms remains a significant challenge. Researchers from the University of Sao Paulo have developed a novel approach to fault detection in wind turbines, combining expert-driven diagnostic knowledge with advanced data-driven modeling. This hybrid method could revolutionize predictive maintenance in the wind energy sector, offering substantial commercial benefits.
The research, led by Welker Facchini Nogueira from the Department of Mechatronics and Mechanical Systems Engineering at the University of Sao Paulo, integrates autoencoder-based neural networks with Failure Mode and Symptoms Analysis (FMSA). This fusion of methodologies enhances anomaly detection, feature selection, and fault localization, providing a more robust and interpretable system.
“Our approach leverages the strengths of both expert-driven and data-driven methodologies,” Nogueira explains. “By combining FMSA with autoencoder-based neural networks, we can improve the precision and interpretability of fault detection in wind turbines.”
The methodology involves five main stages: identifying failure modes and symptoms using FMSA, acquiring and preprocessing SCADA monitoring data, developing dedicated autoencoder models trained on healthy operational data, implementing an anomaly detection strategy based on reconstruction error, and evaluating performance metrics. The approach adopts a fault-specific modeling strategy, where each turbine and failure mode is associated with a customized autoencoder.
The system was first validated using simulated data with induced faults, achieving 99% classification accuracy. It was then applied to real-world SCADA data from wind turbines operated by EDP, successfully identifying degradations in components such as the transformer, gearbox, generator, and hydraulic group up to 60 days before reported failures.
“This hybrid approach demonstrates the potential to support predictive maintenance in complex industrial environments,” Nogueira states. “By detecting faults early, we can prevent costly breakdowns and extend the lifespan of wind turbines, ultimately reducing the levelized cost of energy.”
The research, published in the journal Sensors, highlights the potential of this hybrid approach to shape future developments in the field. By improving the reliability and maintainability of wind farms, this methodology could significantly enhance the commercial viability of wind power, making it an even more attractive option in the global shift toward clean energy.
As the energy sector continues to evolve, the integration of advanced technologies like autoencoder-based neural networks and FMSA could pave the way for more efficient and effective predictive maintenance strategies. This research not only advances the field of wind energy but also sets a precedent for the application of hybrid methodologies in other industrial sectors.