Instrumentation & Measurement Magazine 24-3 - 48

modes corresponding to a defect on the fixed bearing race are
stationary. However, the proposed EKD technique using the
denoising filter (Fig. 5f) can provide the best diagnostic result
with the highest spectral magnitude plus a clear second harmonic due to its feature enhancement effects. Although the
FFT can also recognize the characteristic frequency fod = 91.57
Hz, its spectrum contains extra frequency components that
could degrade diagnostic accuracy, especially in automatic
bearing health condition monitoring.

Inner Race Fault Detection
As stated before, the resonance modes caused by impacts due
to a defect on a rotating bearing component would be nonstationary in nature, especially considering slippage among
bearing components. Fig. 6a, Fig. 6b, and Fig. 6c show the fault
detection results for a bearing with an inner race fault, with
characteristic frequency fid = 148.43 Hz. The FFT fails to predict the bearing defect in this case, as it is not suitable to process
nonstationary signals. Comparing Fig. 6b and Fig. 6c, it is clear
that denoising is necessary for bearing fault detection to improve the SNR. Although the EKD without using the denoising
filter can also recognize this bearing fault (Fig. 6b), the dominant spectral component is the shaft speed (30 Hz), which may
generate false alarms in automatic monitoring applications.

Rolling Element Fault Detection
The detection of fault on rolling elements may be the most
challenging task in bearing health condition monitoring. This
is because: the resonance impulse signatures generated by a
rolling element defect are nonstationary; and the impacts are
random since the defect location may not always strike the
bearing races over the load zone. Fig. 6d, Fig. 6e, and Fig. 6f
show the processing results for a bearing with a rolling element defect, with characteristic frequency fed = 119.52 Hz from
(6). The EKD using the denoising filter in Fig. 6f is the only
technique that can recognize bearing defect in this case, even
though the defect characteristic frequency component does
not dominate the spectrum.

Conclusion and Discussions
Rolling element bearings are commonly used in rotating
machines. In addition, most rotating machinery imperfections are related to bearing defects, for example, up to 50% of
large machinery equipment and up to 75% of small and medium machines. Therefore, reliable bearing fault detection
techniques and tools are critically needed in machinery predictive maintenance applications. Bearing defect signals can
be either stationary or nonstationary, depending on the defect location on the bearing. It is relatively easy to analyze the

Fig. 6. (a), (b), (c): The processing results for a bearing with an inner race defect (fid = 148.43 Hz). (d), (e), (f): The processing results for a bearing with a rolling
element defect (fed = 119.52 Hz). (a), (d): Using FFT; (b), (e): Using EKD without the denoising filter; (c), (f): using EKD with the denoising filter.
48

IEEE Instrumentation & Measurement Magazine

May 2021

Instrumentation & Measurement Magazine 24-3

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 24-3