Windowed Correlated Kurtosis for Isolation and Identification of Compound Faults in Locomotive Bearings

Abstract

Compound faults commonly occur on the rolling bearings. The faults isolation and identification of locomotive bearings are significantly important for bearing health management while difficult in engineering applications. In this paper, a method termed windowed correlated kurtosis (WCK) is presented to successively isolate each impulsive fault mode from compound-fault signals and identify the defects number. To apply WCK for analyzing noise-rich, experiment measurements, a technique frame is proposed which mainly contains two steps: frequency filtering; and WCK based faults isolation. During the filtering procedure, flexible wavelet frames such as tunable Q-factor wavelet transform and flexible analytical wavelet transform which possess arbitrary time-frequency partition property are suggested to adaptively filter the raw vibration measurements such that the signal-to-noise ratio could be enhanced in the filtered signal. Further, WCK is conducted on the filtered signal. The fault modes will be successively isolated in the WCK outputs and the fault signature would be captured on the envelope spectrum. The effectiveness of the proposed technique is validated via analysis of experimental data measured from damaged locomotive bearings subjected to multiple defects on the outer race and roller elements.