An innovative hybrid approach for better detection of bearing faults in highly noisy environments

Abstract

Rolling element bearings (REBs) are critical components in rotating machinery, and timely detection of faults in these bearings can prevent catastrophic failure and costly downtime. Vibration signals collected from REBs contain important information regarding the bearing health states. A localized defect in a REB produces periodic impulses in the vibration signals at relevant bearing characteristic frequencies (BCFs) that are commonly used for bearing condition monitoring. Envelope analysis that has been a powerful technique in detecting BCFs in signals, however, poses a significant challenge as it requires the selection of an effective band-pass filtering region due to the variable nature of resonant frequencies in bearing systems. A novel hybrid method for bearing fault diagnosis based on Cepstrum Pre-Whitening (CPW) and high-pass filtering has been developed to overcome this shortcoming in envelope analysis and has shown to be a powerful technique in accurately detecting early bearing faults on a range of machines under different operating conditions. In contrast to the existing signal processing methods, the new hybrid method eliminates the need for manual parameter selection or optimization algorithms, such as filtering bands selection in enveloping analysis. Instead, it requires inputs including bearing geometries and its operating conditions such as shaft rotating speed for bearing fault diagnosis. This enhances the practicality and accuracy of bearing fault diagnosis in real-world scenarios, especially when data is limited. This study evaluates the effectiveness of this hybrid method for bearing fault diagnosis in the highly noisy environments under which machine is often operated in real applications. To achieve this purpose, firstly, numerical simulations of bearing vibration signals have been conducted, wherein Gaussian white noise with varying levels of signal-to-noise ratio (SNR) is DOI: 10.1784/cm2023.3f4 added to the simulated signals. Secondly, the effectiveness of the method is tested on vibration data collected from an experimental run-to-failure bearing dataset, which contains significant amount of background noise in the early stage of the fault progression. The results from this work have shown that the new hybrid method achieves high accuracy in diagnosing fault features in the simulated signals with a SNR of down to ‐ 13 dB for outer ring fault and ‐ 15 dB for inner ring and ball faults, much more effective than other methods in the literature. It has also shown to detect bearing faults at very early stages from a run-to-failure test.