Máquina de vetores de suporte e técnicas estatísticas para previsão e classificação de falha de turbinas eólicas

Abstract

Wind energy is dominant technology among the generation of renewable energies in Brazil and due to the high growth in recent years there is a need to develop accurate fault diagnosis techniques in wind turbines to reduce downtime due to defects or unnecessary maintenance. Most traditional fault identification techniques are not able to detect defects caused by asymmetry / unbalance in mass or pitch angle on the blades, common faults in wind turbines. Methods such as vibration analysis, which is competent, require the installation of extra sensors in places that may be difficult to access. A new approach would be the diagnosis based on electrical signals from the wind turbine generators, as they prove to be a more reliable and economical option as they do not require the installation of vibration sensors. So, a structure with TurbSim / FAST / Simulink was used to simulate electrical signals generated from a 1.5 MW wind turbine for different scenarios of wind influx and mass unbalance parameters in the blades and pitch angles, machine learning algorithms and statistical tools such as PCA, LDA, EPVM and FFT were applied to the simulation data for fault identification. In the identification of failure conditions, at best, 100.00% accuracy was obtained with electrical signals from the wind turbine generator.