Rolling bearing fault diagnosis based on higher-order cumulants and BP neural network

Abstract

Based on the fact that the rolling bearing fault vibration signals are susceptible to Gauss noise, a fault diagnosis of rolling bearing method using higher-order cumulants and back propagation (BP) neural network is proposed. In this paper, the higher-order statistics of the vibration signals are calculated as feature vectors, including the third-order cumulant and the fourth-order cumulant as well as the second-order cumulant. And a BP neural network is trained to identify the bearing fault by using those features. The effectiveness of the proposed method is verified by four types of rolling bearing, namely ball fault, inner raceway fault, outer raceway fault, and normal bearing. The experimental results show cumulants based fault features have perfect separation. Except the training and test diagnostic accuracy of ball fault are high as 98.75 % and 96.67%, classification accuracies of other faults rate are 100%.