Rolling Element Bearing Fault Identification using Vibration Data

Abstract

Early fault detection and diagnosis of rolling element bearing is of paramount importance in wind turbines as it contributes to around 70% of gearbox and 21%–70% of generator failure. When a rolling element bearing strikes a local fault in the inner or outer race, a shock (high frequency) is introduced, and repetitive impact occurs due to continuous rotation. Extracting the fault-sensitive repetitive impact frequency component from the measured signal containing multiple frequencies (discrete gear and shaft frequency, bearing fault frequency and high-frequency noise) is challenging. This paper presents two vibration techniques based on enhanced envelope analysis and blind deconvolution technique for bearing fault identification. The improved envelope analysis diagnosis bearing faults using the three-step process of removing gear and shaft frequency components by auto-regression model, followed by spectral kurtosis to extract fault-sensitive features and envelope analysis to identify bearing faults. On the contrary, the enhanced blind deconvolution extracts the fault-sensitive component by finding the best inverse finite impulse response filter from the measured vibration signal by adaptively demodulating resonance bands due to repetitive impact and reducing the periodic noise component in a single step. The application of the two bearing fault diagnosis techniques and their comparative study has been demonstrated through numerical simulations and two industrial testrig bearing benchmark datasets. Investigations concluded that both methods extract the transient impulse features due to bearing fault; however, the enhanced blind deconvolution technique outperforms the envelope analysis in the case of measured vibration signal with outliers.