A Wind Turbine Feature Extraction Method Based on Kernel-Domain Spectrum

摘要

In this paper, a new kernel-domain spectrum based method is proposed to analyze the non-Gaussian and nonlinear characteristics of vibration signal for fault diagnosis of wind turbine. The proposed approach consists of the following key steps. First, the raw vibration signal measured from the wind turbine is divided into groups for data pre-process. The bispectrum method is then applied to the group data analysis and accordingly two-dimension data matrices are obtained. The baseline kernel templates, which will be utilized for comparison and classification, are then built up by a binarization processing. The fault classification can be achieved by a comparison between baseline kernel templates and test kernel templates, and the latter templates are built by using the same method on the test data. This kernel-domain spectrum based method can not only distinguish fault types, but also address the challenges associated with smaller faults. The testing results from the rolling-element bearing fault diagnosis experiment proved the effectiveness and viability of the proposed method.