Rolling Bearing Fault Diagnosis Based on Wavelet Packet and RBF Neural Network

Based upon wavelet packet analysis and radial basis function (RBF) neural network, a method for the fault diagnosis of roller bearings is proposed in this paper. Firstly, wavelet package was used to decompose vibration time signals of bearing to extract the characteristic values-wavelet packet energy, and features were input into the RBF NN. After training, the RBF NN could be used to identify the roll bearing fault patterns. Three typical bearing faults such as inner race fault, outer race fault and rolling element fault were studied. The results showed that the method of RBF NN with wavelet packet could not only detect the exiting of the fault in bearings, but also effectively identify the fault patterns. Therefore, the presented modular RBF networks are quite suitable for the large sample problems. And also, compared the effects of the daubechies (db) 8 and the symlet (sym) 8 on the analyzing roller bearings signals, the former is better .