L-Kurtosis and its application for fault detection of rolling element bearings

Kurtosis is a useful indicator of fault detection. However, the intrinsic drawback of kurtosis makes it very susceptible to the outliers. Kurtosis is hence not an ideal feature in vibration signal analysis. L-kurtosis is a fourth-order L-moment, which is similar to kurtosis and easy to recognize the impulse but is not like kurtosis to be sensitive to the outliers. This paper introduces the definition, properties of L-kurtosis and analyzes the reason that why L-kurtosis is robust to the outliers. Through inferring some typical distributions of L-kurtosis, this paper proposes that L-kurtosis value 0.1226 may replace kurtosis value 3 as a threshold in fault diagnosis. After that, the simulated fault data additive Gaussian noise and interference validated L-kurtosis outperforms over kurtosis. The following experiments about rolling bearing faults are further verified its effectiveness in selecting the proper bandwidth and identifying the fault characteristics. As a result, L-kurtosis is an ideal feature in vibration signal analysis.