Luo Y., Daley S.

A number of recent studies have evaluated the efficacy of various feature extraction methods in the detection of rotor cracks using vibration signatures for machines undergoing transient operation. However, in many practical applications, rotating machines run at steady state for most of their operational life. This paper describes an investigation of feature extraction based on data obtained from a rotor running at steady state which is considered to represent a more realistic assessment of the feasibility of online crack detection. A systematic comparative study of a number of widely used feature extraction methods for crack detection on rotors based on simulated vibration data is considered and compared with earlier work by the authors that concentrated solely on transient operation. The simulated cracked rotor utilises a simple hinge model together with a wide range of different crack depths which are represented by a change in stiffness. Measurement noise is also included for comparing the fault sensitivity of the various detection methods. Based on the normal range of operating speed for rotating machines, the analysis is divided into two groups; a sub-critical speed range (ranging from 0 to 1st machine critical speed) and a super-critical speed range (ranging between 1st and 2nd machine critical speed). Rather than using visual discrimination or sophisticated comparison methodologies, simple statistics are adopted based on the extracted features. The main purpose of the study was to determine both the optimal operational speed and the optimal feature extraction method for crack detection. For the noise free case it was found that the greatest sensitivity to cracks occurred at the lower end of each operational speed range. For the majority of cases the features extracted from the bispectrum provide the highest fault sensitivity.