Prediction of the vibration response due to defects on rolling element bearings requires having an accurate representative vibration model. In this paper, a new model for vibration generation in the rolling element bearings has been introduced. The proposed model assumes a stochastic source of vibration excitation, which is produced as a result of metallic contact between bearing elements during rolling. This model explains high frequency vibration in the acceleration spectrum clearly. When a defect grows in the bearing, the roughness of the contacting surfaces increases locally and stochastic excitation becomes stronger in the defective area. The increased vibration level at the defective area is a good indicator of bearing faults. A numerical simulation of the proposed model was validated with experimental results.

Selecting appropriate features from the vibration condition monitoring data of ball-bearings is one of the main challenges in the application of data-driven methods for remaining useful life prediction purpose. In this article, a new feature based on the high-frequency vibration of ball-bearings is proposed. The feed forward neural network will be used for training and prediction. The experimental data of the bearing accelerated life in the PROGNOSTIA test (published in PHM 2012 IEEE conference) are used to verify the method. The results obtained by applying new features are compared with those of two popular features in the time domain (RMS and kurtosis) for prognostic purpose. Applying the proposed feature shows more accurate estimation of the bearings’ remaining useful life.