Recognition of Rolling Bearing Based on EMD and SOM Neural Network

Fault recognition of rolling bearings is the basis of condition-based maintenance. Aiming at the non-stationarity and non-linearity of vibration signals emitted from defective bearings, a fault recognition method is proposed based on Empirical Mode Decomposition (EMD) and Self-Organizing Feature Maps (SOM) neural networks. Vibration signals are decomposed into a collection of IMFs (Intrinsic Mode Functions) by EMD, and then the energy features extracted from IMFs containing fault information are treated as input of SOM neural network. Various bearing health conditions involving different fault types and severity levels are identified by the SOM. Experimental results verified the effectiveness of the proposed method.