Fault Diagnosis of Wind Turbines Gearbox Based on SOFM Neural Network

Abstract

In view of the shortcomings of wind turbines gearbox fault diagnosis technology, this paper presents a diagnosis method based on self-organizing feature mapping (SOFM) neural network. First denoised the vibration signals of a wind turbine gearbox in its normal state, wear fault and tooth breakage through wavelet analysis method. Then five fault feature indexes in time domain and frequency domain were taken as input eigenvectors to train the network. And diagnosed the fault type according to the location of output neurons on output layer. At last a fault diagnostic model based on SOFM neural network was built. In order to test its diagnostic ability, the built model was used to diagnose the measured data of wind turbine gearboxes of a wind farm in northern China. The simulation results show that the built model can judge the fault type according to the location of winning neurons in the competing layer. And its diagnosis accuracy is high; its convergence speed is fast and its generalization ability is also good. It is indicated that the established network model can effectively diagnose gearbox fault.