Extented Intelligent Recognition of Rolling Bearing Early Faults Using Multiscale Permutation Entropy

Abstract

Considering the diversity, complexity and uncertainty existing in bearing vibrations, an extended intelligent identification paradigm for bearing faults was proposed based on multiscale permutation entropy (MPE) and extension theory. MPE can reflect the random degree and detect the dynamic mutation of time series over subsequent scales, while extension theory provides an approach to address the extensibility and regularity of complicated problems. In the present paradigm, MPE was employed to compute the entropies over multiple scales as an original feature vector to represent bearing vibrations, which were then graded using Fisher ratio to choose the most informative features. The chosen features were exploited to determine the classical domain and joint domain of matter elements associated with various bearing health conditions. Bearing fault pattern was assigned to the one with maximum dependence degree among the afore-constructed matter elements. An experiment was conducted on an electrical motor involving four bearing conditions including normal, inner race, outer race and rolling element faults. The test was repeated 100 times with an averaged rate of 92.2% by the proposed method which outperforms the method using multiscale sample entropy and extension theory.