Fault Diagnosis of Bearings Based on Wpd and Rbf Neural Networks

Conventional signal processing techniques usually result in false information when they are applied to the ship mechanical fault signals, because the ship mechanical faults by nature are non-stationary and transient events. Wavelet Packet Decomposition (WPD) is a time– frequency domain technique that can be applied to non-stationary process perfectly. RBF neural network behave better than BP neural network in approximation ability, classification ability and learning speed. A new fault diagnosis method based on WPD method and RBF neural network is presented. With the method, the rolling element bearings vibration signals are decomposed into several frequency bands from high to low with WPD, trained and configured networks with the energy characteristics of frequency bands are used to detect the novelties or anomalies of faulty signals. The proposed method is applied to the fault diagnosis of rolling element bearings, and the entire 80 test results could correctly identify the bearing faults. The results show that the combination of WPD and RBF neural networks could reliably separate different fault conditions.