The Squared Envelope Spectrum (SES) and the Improved Envelope Spectrum (IES) are some of the most powerful representations for diagnosing rotating machine components such as bearings [1, 2]. However, to ensure that the SES and the IES representations are rich with diagnostic information, it is necessary to confine the SES and the IES calculations in order to narrow the spectral frequency bands that contain the required diagnostic information [2]. Various methods have been proposed in order to automatically identify the appropriate frequency band for fault detection and diagnosis such as the kurtogram [3], the infogram [4] and the sparsogram [5]. This is usually performed by decomposing the signal into a range of bandlimited signals and selecting the frequency band that maximises a metric, i.e. the kurtosis [3].

Machines such as wind turbine gearboxes inherently operate under time-varying operating conditions and could operate in an environment with transients unrelated to the machine component being monitored (e.g. a spurious impulse). These phenomena could adversely affect the performance of the frequency band identification techniques. For example, the presence of transients unrelated to the machine condition can lead to the identification of a suboptimal or wrong frequency band when metrics such as the kurtosis [4] are used.

Hence, in this paper a new Frequency Band Identification Tool (FBIT) is proposed which is suited for signals acquired under non-stationary operating conditions. This FBIT is used to automatically identify the frequency band that contains the signal of interest and the most information related to a specific damage mode of a machine component (e.g. ball pass cyclic order of a bearing and its harmonics).

In this FBIT, the signal is also paved into a spectral frequency-frequency resolution plane, but instead of calculating a metric of the signals based on multiple short-time Fourier transforms or a wavelet packet decomposition, a metric is calculated from the data based on multiple order-frequency spectral coherences. This is motivated by the fact that the order-frequency spectral coherence is a very powerful representation for detecting fault signatures of bearings and gears under time-varying operating conditions and it preserves the angle-time cyclostationary properties of the fault signatures as well. Also, a different metric is used compared to conventional approaches. Instead of calculating a metric of the whole narrowband signal (e.g. kurtosis) which may be sensitive to unrelated outliers (e.g. spurious impulse), a metric is calculated which represents the strength of the specific cyclic component in the signal and by optimising also the window resolution of the spectral coherence. This allows the selection of the optimum frequency band for the identification of the cyclic component under investigation (e.g. for identifying outer race bearing damage).

The proposed FBIT is investigated on numerical bearing data, generated from a phenomenological gearbox model under time-varying speed conditions, as well as on experimental gearbox data, acquired under varying speed and load conditions.

Keywords

Gearbox fault diagnosis; Order-Frequency Spectral Coherence; Frequency Band Identification

References

[1] Abboud, D. et al., 2017. Envelope analysis of rotating machine vibrations in variable speed conditions: A comprehensive treatment. Mechanical Systems and Signal Processing, 84, pp.200–226.

[2] Abboud, D. & Antoni, J., 2017. Order-frequency analysis of machine signals. Mechanical Systems and Signal Processing, 87(October 2016), pp.229–258.

[3] Antoni, J., 2007. Fast computation of the kurtogram for the detection of transient faults. Mechanical Systems and Signal Processing, 21(1), pp.108–124.

[4] Antoni, J., 2016. The infogram: Entropic evidence of the signature of repetitive transients. Mechanical Systems and Signal Processing, 74, pp.73–94.

[5] Tse, P.W. & Wang, D., 2013. The design of a new sparsogram for fast bearing fault diagnosis: Part 1 of the two related manuscripts that have a joint title as “two automatic vibration-based fault diagnostic methods using the novel sparsity measurement - Parts 1 and 2.” Mechanical Systems and Signal Processing, 40(2), pp.499–519.