Abstract:

According to the nonlinear and non-Guassian characteristics of vibration signals of rolling bearings,a novel fault identification method based on the bispectrum distribution feature of auto-regressive moving average ( ARMA) model and on the cluster analysis of fuzzy c-means ( FCM) method is proposed. In this method,first,original vibration signals are modulated via the empirical mode decomposition ( EMD),and an ARMA model of principal
signal components is established. Then,a bispectrum estimation of the ARMA model is implemented. Finally,the binary images extracted from the bispectrum distribution are taken as the feature vectors and are used to construct a classifier of the class templates and the smallest-distance templates via the FCM clustering,thus implementing the fault identification successfully. Application results in the fault diagnosis of rolling bearings demonstrate that the proposed method is effective because it can accurately determine the actual conditions of rolling bearings.

Key words: fault identification, bearing, ARMA model, bispectrum, fuzzy c-means