Bispectral feature analysis and diagnosis for bearing failure of direct-drive wind turbine

Vibration measurements on a large direct-drive wind turbine are carried out to detect the fault of the rolling element bearing on the rear part of the main shaft. The main purpose of this paper is to find an appropriate method for weak fault feature extraction and fault recognition from vibration signals under strong interference. We propose to use higher order statistics characteristics of vibration signals for differentiating normal condition from fault condition. Firstly the non-Gaussian intensity in different area in vibration signal bispectrum are extracted as the feature values, these multi-dimensional features are compressed by means of principal component analysis (PCA) to obtain some lower dimensional principal component features with better discrimination for different running condition. Analysis results show that the proposed method of feature extraction with the non-Gaussian intensity characteristic of the bispectrum is very sensitive to differentiate the normal running condition from failure ones and very clear to identify the bearing fault of wind turbine.