Automatic fault recognition in mechanical components using coupled artificial neural networks

Vibration monitoring is based on the principle that all systems produce vibration which may be analyzed to gather operational performance information. In rolling element bearings (REB), the components investigated in this article, deterioration is manifested in loss of fragments of metal of the raceways or the rolling elements. Each time a rolling element passes over a defect, an impulse of vibration is generated. Impulses are generally of very short duration, generate responses in a wide range of frequencies, and are related to faults and severity. All vibration monitoring techniques are based fundamentally on the identification and quantification of these impulses. Generally when a machine is operating properly, overall vibration is small and constant (although “beat frequencies” may appear); however, when faults develop or when some of the dynamic processes in the machine change, the vibration spectrum also changes. The frequency spectrum, generated using fast Fourier transforms, is very useful because in the frequency domain low level signals are not masked by the presence of high level signals, and because the Fourier spectrum (phase and amplitude) describes completely the vibration of the component