Abstract:

Research is conducted on the characteristics and diagnostic methods of early dynamic eccentricity and early static eccentricity faults in the rotor of permanent magnet synchronous wind turbines.The early dynamic eccentricity and early static eccentricity models for permanent magnet synchronous wind turbines are established using Ansys,and a fault diagnosis and classification method based on CNN-LSTM is proposed.By analyzing the three-phase current and Welch power spectrum data of the stator of a permanent magnet synchronous wind turbine generator,the generator’s current status can be judged whether it is normal dynamic eccentricity trend or normal static eccentricity trend.Then,the different fault levels are classified using no-load electromotive force.Finally,the fault diagnosis and classification tasks in the neural network model are completed.The proposed method greatly reduces equipment maintenance costs and can accurately and quickly identify early rotor eccentricity faults.

Key words: convolutional neural network (CNN), long short term memory (LSTM) network, fault diagnosis, feature extraction