Bearing diagnostics based on spectral kurtosis

Bearings are a critical component of a rotary engine. Bearing faults caus a decrease in engine performance and has the potential to stop a production process. Therefore, monitoring and diagnosis of bearing conditions is very important. Envelope analysis is a reliable method used to diagnose bearings. The selection of demodulation band in envelope analysis is often done by comparing the healthy bearing spectrum with the faulty bearing spectrum to determine the resonance frequency at which the transient signal occurs due to the fault. However, information on the spectrum of healthy bearings is not always practically available in the field. Spectral kurtosis (SK) is known as a technique that can determine the modulated band without the condition of historical information. Kurtosis is a statistical parameter whose value is close to zero for gaussian noise and stationary signals, but its value is large for impulsive signals with short transient properties such as those produced by bearing faults. The paper proposes the application of SK in determining the optimum resonance band on the analysis envelope for diagnosing bearings. SK analysis parameters (i.e. window length) were explored to obtain optimal results. The diagnostic results are presented using the actual signal bearing the inner race fault.