Detection and localization of fatigue-induced transverse crack in a rotor shaft using principal component analysis

Rotor shafts subjected to severe operating stresses are prone to develop transverse fatigue cracks at the localized stress raisers. Therefore, the ability to identify and locate the incipient fatigue crack is imperative in order to avoid catastrophic failure. The literature on rotor crack detection discussed the importance of monitoring the steady-state 1X, 2X and 3X harmonic response components of rotors. However, the other rotor faults such as misalignment and unbalance, exhibit similar symptoms. Thus, the main aim is to develop new independent fault-related features which measure the driving principle governing the behaviour of various rotor faults. In this article, the application of principal component analysis-based statistical pattern analysis, as a tool for early detection and localization of fatigue-induced transverse crack in a rotor shaft is investigated. To perform this study, accelerated fatigue experiments are conducted on a customized setup. This developed test rig is novel and unique by itself that facilitates generating a fatigue crack in a shaft, under conditions that mimic a real in-service loading environment of industrial rotors. Unlike conventional methods, noise in the acquired vibration and strain data is denoised via classical principal component analysis method. Time- and frequency-domain statistical features extracted from different vibration and strain sensor signals are used for this study. Damage indices such as Hotelling'sT(2)-statistic andQ-index are used to detect the presence of the crack. It is observed that irrespective of the sensor location, damage index such as Q-statistic of all the sensors is very effective to detect the presence and time of incipient crack. Partial decomposition contributions method is found to be very effective in identifying the location of the crack. This article provides the most significant vibration-based statistical features, which are sensitive to shaft transverse cracks, for different sensor types and their mounting location. Finally, a new fused health indicator which is highly sensitive to the presence of rotor shaft crack is defined and is found successful when applied to a new experimental data.

Automated summary

This article investigates the application of principal component analysis-based statistical pattern analysis for early detection and localization of fatigue-induced transverse cracks in rotor shafts. The study utilizes accelerated fatigue experiments on a customized setup and denoises noise in acquired vibration and strain data using classical principal component analysis. Time- and frequency-domain statistical features extracted from different sensor signals contribute to the development of a new fused health indicator sensitive to rotor shaft cracks.