Short-Circuit Phase Detection in Wind Turbines Using Subtle Neural Network Learning

Identifying the phase in which a short-circuit occurs in wind turbines is challenging because of subtle variations in the generator’s operating characteristics, such as changes in current, which are often imperceptible. Even when the short-circuit is detected, determining the affected phase is difficult, as short circuits in different phases may exhibit similar behaviors. Interturns short-circuits in the generator windings are even more challenging to detect, as they do not cause immediate damage but can reduce the wind turbine’s lifespan. This study presents a solution using a modular neural network topology based on decision-making and council models. The model was trained only with simulation data and tested using data collected from a generator in a controlled environment. The topology achieved 98% accuracy. The model’s generalization capability, combined with the use of only current data, enables the development of a prototype for future validation of the results in real-world wind turbines. More accurate fault classification results in more efficient monitoring systems and impacts the machine maintenance philosophy. The main goal is to contribute to reducing downtime and increasing industry revenues.