New paper: Improving wind farm reliability using generative domain mapping

In our new article in Energy and AI, we address wind farm condition monitoring from a new angle, using generative domain mapping across wind turbines. In this work, led by PhD student Stefan Jonas, we show that normal behaviour models for wind turbines trained on limited data can lead to unreliable condition monitoring. To address this, we introduce a domain mapping method that improves anomaly detection and yields more accurate anomaly scores. Using a CycleGAN model, we map SCADA data from data-scarce wind turbines to data-rich turbines while preserving turbine states, including anomalies. Our presented domain mapping approach can make differing wind turbines resemble each other. It achieves earlier fault detection due to more accurate anomaly scores.

Paper [Link]

Code [Link]