A promising new tool for fault diagnosis of railway wheelset bearings: SSO-based Kurtogram.

A promising method is proposed systematically to select an accurate resonance frequency band and separate refined resonance response from periodic excitation in this study. This work expanded the short-time Fourier transform (STFT)- and wavelet transform (WT)-based Kurtograms and developed a hybrid signal separation operator (SSO)-spectral kurtosis computational scheme to implement Kurtogram by introducing the SSO method-SSO-based Kurtogram. The ability to accurately extract the refined resonance frequency band of SSO greatly improves its adaptivity for engineering applications. The effectiveness of the SSO-based Kurtogram is studied by using a bearing fault simulation signal, and the influence of window function on the detection effect of the proposed method is explored. Furthermore the validity of the SSO-based Kurtogram for bearing fault detection is verified by a set of railway wheelset-bearing experiments on the wheelset running-in testbed bench. Experimental results show that the SSO-based Kurtogram performs highly in detecting various kinds of single and compound faults of bearings. Compared with the WT- and STFT-based Kurtogram, the proposed method has obvious advantages in terms of effectiveness and visual inspection ability. In engineering practice, a railway wheelset-bearing-fault experiment on an in-service high-speed train in the real world is taken as a case study, which makes the verification of SSO-based Kurtogram more convincing and demonstrates the practical engineering value of the proposed method. The results show that in case of equal effectiveness, SSO-based Kurtogram has an absolute advantage in the visual inspection ability, embodied in eliminating other vibrations unrelated to the target fault and making the fault feature frequency and its harmonics remarkable.