Simulation-Based Neural Network for Robust Short-Circuit Detection in Wind Turbines

Fault detection in wind turbines is a widely discussed topic because of advancements in the energy transition. Short-circuit faults between turns in the generator windings are challenging to detect. These faults do not compromise turbine operation, but can reduce their lifespan. Many technologies currently available in the market cannot detect this defect. This study presents a neural network algorithm to address this issue. Its simple architecture delivers satisfactory performance for certain types of generators, and, importantly, the neural network is trained solely with simulation data. This eliminates the need for time and resources to acquire real-world data from wind turbines. The results are validated using data from two experimental platforms with different configurations, demonstrating the algorithm’s generalisation capability.