Gearbox Diagnosis Based on the Spectral Kurtosis and Adaptive Filtering

Gearbox diagnosis, specifically diagnosis of bearings and gears, is traditionally done using envelope demodulation approach. The spectral kurtosis (SK) is commonly used to identify the frequency band for demodulation, which is related to the structural resonances excited by a series of fault-induced impulses. The diagnosis technology proposed in this paper follows the traditional use of the SK and applies the optimal denoising (Wiener) filter based on the SK to relatively short vibration segments. The filtered signal, called the SK-residual, is used to extract the diagnostic features in terms of the squared envelope, which is subjected to decision making process by k-means and k-nearest neighbours. The originalities of the proposed technology are robustness to fluctuating operating conditions and random slippage in case of bearing damage diagnosis, due to processing of relatively short signal realizations by frequently adapted SK-based Wiener filter. In addition, instead of performing classical envelope spectrum analysis, the approach uses analysis of the squared envelope in time domain to achieve reliable damage diagnosis. The resolution for SK estimation, optimal threshold for filtering, and damage diagnosis processes are discussed in the paper. The technology is experimentally tested for bearing inner race defect, and simulated gear vibration including 15% pitting damage size.