A fuzzy clustering-based binary threshold bispectrum estimation approach

Abstract

A fuzzy clustering bispectrum estimation approach is proposed in this paper and applied on the rolling element bearing fault recognition. The method combines the basic higher order spectrum theory and fuzzy clustering technique in data mining. At first, all the bispectrum estimation results of the training samples and test samples are taken binarization threshold processing and turned into binary feature images. Then, the binary feature images of the training samples are used to construct object templates including kernel images and domain images. Every fault category has one object templates. At last, by calculating the distances between test samples’ binary feature images and the different object templates, the object classification and pattern recognition can be effectively accomplished. Bearing is the most important and much easier to be damaged component in rotating machinery. Furthermore, there exist large amounts of noise jamming and nonlinear coupling components in bearing vibration signals. Higher Order Cumulants, which can quantitatively describe the nonlinear characteristic signals with close relationship between the mechanical faults, are introduced in this paper to de-noise the raw bearing vibration signals and obtain the bispectrum estimation pictures. At last, the rolling bearing fault diagnosis experiment results showed that the classification was completely correct.