Vibration signals of faulty rolling bearing contain a large amount of information about the bearing operating status. However

it is difficult to extract the fault features completely because of its characteristics of nonlinearity and non-stationarity

which leads to a problem of relatively low fault identification rate of current fault diagnosis methods based on pattern recognition. In order to improve the accuracy of rolling bearing fault diagnosis

this paper proposes a fault diagnosis method of rolling bearing

which adopts ensemble empirical mode decomposition and permutation entropy (EEMD-PE) to extract the fault features coupled with multiclass relevance vector machine (M-RVM) to achieve the goal of fault classification. Firstly

the vibration signal of faulty rolling bearing decomposes into a series of intrinsic mode functions (IMFs) by using the adaptive decomposition ability of nonlinear and non-stationary signals. Afterwards

the fault features contained in IMFs are extracted by permutation entropy

and the features constitute the feauture vector. Finally

EEMD-PE method is used to extract the fault feaures of training sample set under different fault conditions. The M-RVM classifier is trained by using feature vector set

and the multiple fault identification is implemented in the form of probability output. The experimental results show that EEMD-PE feature extraction method can effectively extract fault features of rolling bearing vibration signal

M-RVM can identify the fault feature contained in rolling bearing vibration signals. Compared with the existing bearing fault diagnosis methods

this method can improve the fault identification rate reaching up to 99.58%.