Rolling Bearing Fault Detection Based on the Teager Energy Operator and Elman Neural Network

Abstract

This paper presents an approach to bearing fault diagnosis based on the Teager energy operator (TEO) and Elman neural network. The TEO can estimate the total mechanical energy required to generate signals, thereby resulting in good time resolution and self-adaptability to transient signals. These attributes reflect the advantage of detecting signal impact characteristics. To detect the impact characteristics of the vibration signals of bearing faults, we used the TEO to extract the cyclical impact caused by bearing failure and applied the wavelet packet to reduce the noise of the Teager energy signal. This approach also enabled the extraction of bearing fault feature frequencies, which were identified using the fast Fourier transform of Teager energy. The feature frequencies of the inner and outer faults, as well as the ratio of resonance frequency band energy to total energy in the Teager spectrum, were extracted as feature vectors. In order to avoid a frequency leak error, the weighted Teager spectrum around the fault frequency was extracted as feature vector. These vectors were then used to train the Elman neural network and improve the robustness of the diagnostic algorithm. Experimental results indicate that the proposed approach effectively detects bearing faults under variable conditions.