Using bispectral distribution as a feature for rotating machinery fault diagnosis

The vibration signals of rotating machinery present a strongly non-linear and non-Gaussian behavior, and bispectrum is well suitable to analyze this kind of signals. Due to modulation or smearing, it is hard to extract the accurate frequency-based features from the bispectrum. A bispectral distribution for machinery fault diagnosis is developed in this paper. The binary images extracted from the bispectra are taken as features to construct the target templates, then, the nearest template classifier is constructed to achieve pattern recognition and fault diagnosis. The computing speed of this method is very high because the proposed algorithm just calculates the number of “1”. Finally, roller bearing and gear fault diagnosis are performed as examples, respectively, to verify the feasibility of the proposed method.

► The same roller bearing/gear fault types are of well similarities in their bispectral distribution regions, and vice versa.
► The binary images extracted from the bispectra are taken as features.
► The nearest template classifier is constructed to achieve pattern recognition.
► This diagnostic process just simply calculates the number of “1” which can greatly improve the computing speed.
► Roller bearing and gear fault diagnosis are taken as case study to demonstrate the feasibility and advantages of the proposed method.