Abstract
Given the nonlinear and nonstationary characteristics of wind power generator vibration signals, the empirical wavelet transform (EWT)-based method is proposed to analyze the vibration signal characteristics and to diagnose faults. The adaptive Fourier spectrum segmentation of EWT is analyzed, and an appropriate wavelet filter is constructed to extract different AM-FM mode functions. The Hilbert transform is conducted, and the signal modal component energy is normalized to obtain vibration feature quantities. A probabilistic neural network (PNN) is used to perform the classification and diagnosis of wind power generator vibration faults. The experiments are constructed based on simulation signals and vibration signals before the proposed method and Hilbert-Huang transform (HHT) are used to perform mode decomposition and to analyze the time-frequency energy spectrum. The experimental results showed that the obtained resolution modes by EWT are within the corresponding time domain signal characteristics. The number of mode decomposition layers is less than that of empirical mode decomposition. No characteristics of false modal are observed. The time-frequency energy spectrum diagram can better reflect the characteristics of original vibration signal than the spectrum based on HHT. The PNN-based vibration fault judgment can achieve an accuracy rate of 90 % with limited training samples.

This publication has 25 references indexed in Scilit: