Detecção de desbalanceamento de massa entre pás de aerogerador utilizando sinais elétricos de um gerador síncrono de imãs permanentes aplicados a métodos de aprendizagem de máquina

Abstract

The use of wind power has increased in electricity production due to its cost becoming closer to conventional generation sources. The useful lifetime expectancy of wind turbines are typically of 20 years, however failures are expected much earlier due to components (especially gears, electrical system and blades) failures. Detecting and avoiding harsh operations and conditions of wind turbines are useful to reduce downtimes and catastrophic failures that could lead to greater losses and even a complete shutdown of the turbine. A considerable amount of failures in components can be assign to the rotor imbalance of wind turbines, since the imbalance can cause greater mechanical loads throughout the whole structure. In this paper, machine learning methods are applied in order to detect blade mass imbalance in wind turbines. The data used for such task were obtained from the platform Turbsim/FAST/Simulink, which simulates the dynamics of a 1.5 MW wind turbine, with both normal and blade mass imbalance operations. The electrical quantities obtained from the permanent magnets synchronous generator of the wind turbine allows to estimate the rotational speed of the rotor, and this data is applied to machine learning methods to detect whether there is imbalance or not. The use of coupled aeroelastic numeric simulations and machine learning in this paper presented a feasible method to improve the production of wind power, reducing downtimes, allowing scheduled maintenances and increasing the LCOE of wind power generation.