Fault Diagnosis of Roller Bearing Based on Bispectrum Estimation and Fuzzy Cluster Analysis

Performing bispectrum analysis on the actual measured vibration signals of the roller bearing with different failure modes, it developed that the spectrum distribution regions are similar among the same failure modes, and distinguishable among the different failure modes, thus this character can be used to classify fault types. The binary images extracted from the bispectra are taken as the feature vectors. The fuzzy clustering analysis based on objective function is applied for pattern recognition, which makes use of the binary image to construct a core and a domain representing the common and scope of bispectrum distribution, respectively, then constructs the objective function as the classify to achieve pattern recognition and diagnosis. The roller bearing fault diagnosis is performed as an example to verify the feasibility of the proposed method.