Spectral Correlation Applied to Gear Monitoring in Electrical Power Plants

Abstract

In power plants numerous equipments including gears have to be monitored to prevent damage and non-planned shut-downs. Whereas only visual or aural monitoring by skilled personnel is performed presently, studies are carried out to improve this monitoring by processing vibratory signals provided by sensors. These signals exhibit both amplitude and phase modulations and so can be processed by the spectral correlation approach. We have compared the Wigner-Ville distribution and the spectral correlation and showed that the latter is well adapted to modulated signals analysis. It allows for discriminating between running-in phases, stable phases and aging or damage phases of the component to be monitored. The ability to show in advance signs of degradation through modulation rates changing and also modulation structure variations is demonstrated on real life signals. 1. GEAR MONITORMG M POWER PLANTS Numerous machines in power plants (e.g. rotating machines) contain gears which have to be monitored in order to prevent damage and shut-downs. Today's state of the art in monitoring is based on periodic maintenance. This is not optimal neither technically nor economically. Conditional maintenance will probably replace it, providing that we are able to measure reliable and relevant descriptors of the vibrational behaviour of the component and to derive from them a consistent indicator of degradation. This will finally be used to stop the machine before damage has occurred. The principle of the gear monitoring is to get external signals from accelerometers stucked on the gear box and to analyse them so as to determine whether the gear teeth are damaged or not. correlation which appear to be well suited to signals from rotating machinery. Different descriptors computed from the spectral correlation will be introduced whereas section 4 will explain through a real world experiment on a testbench how they can be used to monitor gear degradation . 2. GEAR SIGNAL MODELISATION A gear is composed o f : a wheel # 1, with N, teeth, rotating at speed F 1, a wheel # 2, with N, teeth, rotating at speed F2. The gear frequency Feng is defined as .the frequency where teeth come into contact with each other : Feng = NI * Fl = N2 * F2 = 1 / Teng. Let us assume that a flaw has appeared on one of the two wheels (typically this flaw is a tooth flaking). This will result in an amplitude modulation combined with a phase modulation [ 11. Hence the vibratory signal produced by the gear rotation can be modeled as follows : sd(t) = p p ( l +op(t))cos(2npF,, +@, + b p ( r ) ) P where : ap(t) = Z A L sin(2niFrt +ai P ) + P J ) P rotation frequency of the faulty wheel. gear frequency . amplitude modulation of the pm harmonic of FmP. phase modulation of the pm harmonic of Fm,. In section 2 we will establish a model for the gear signals before choosing a technique to analyse them. Section 3 introduces the cyclostationarity and the spectral