Abstract: Aiming at the difficulty of recognizing fault pattern of rotating parts in mechanical equipment，a new method for fault diagnosis based on chaos and fuzzy maximum likelihood estimates (FMLE) clustering algorithm is introduced. The non-equilibrium phase change of chaos oscillator is very sensitive to small signal and immune against the random noise and the high frequency signal. The frequency of the weak fault signals is extracted easily, which can be used to cluster analysis as fault feature vectors. Considering that the traditional fuzzy c-means (FCM) clustering algorithm is only suited to spherical-shape distribution dataset, distance norm based on the fuzzy maximum likelihood estimates is introduced, which suits to datasets with different shape and size, density and the different faults in rotating machinery are detected automatically. Results of experimental and engineering test indicated that the method is effective, and the cluster effect based on FMLE clustering is better.

Key words: Fault diagnosis, Fuzzy maximum likelihood estimates clustering algorithm, Intermittent chaos, Max-min close degree