Low-cycle fatigue-initiated cracks may result in failure in slow-rotating equipments. Online monitoring to identify such fault/crack parameters, namely crack size and crack location, would be critical in providing an early warning signal to the operator and would be used in calculating estimate about the remaining safe life of the equipment in operation. In an earlier study, a scaled-down slow-rotating washer drum was constructed to experimentally investigate the vibrations of a cracked rotor and/or drums. Cracks were simulated using the bolt removal method (BRM), and the vibration signals identifying signatures of certain cracks were measured. Thereafter, a 3D finite element model was used to solve the forward analysis of the inverse problem of crack identification. In this paper, the scaled-down experimental setup is introduced to cracks at different locations of the drum/rotor. Vibration signals identifying signatures of such cracks are measured. Since noisy signals, similar patterns of faults, and similar vibration fault signals create particular challenges for feature extraction systems, two techniques for feature extraction are considered and compared in this work. The fast Fourier transform (FFT) of the vibration signals showing variation in amplitude of the harmonics as time progresses are presented for comparison with the full time signal feature extraction. A hybrid particle-swarm artificial Neural Networks (neuro-particle swarm) is used to identify both the crack size and crack location. The hybrid neuro-particle swarm technique is compared with the previously investigated fuzzy genetic algorithms.

This content is only available via PDF.
You do not currently have access to this content.