A Bayesian damage prognosis approach applied to bearing failure

Abstract

Bearings are one of the most important components in rotating machinery, and their failure can cause catastrophic consequences. Conventional damage detection technologies have been successfully applied to make early warning of defect occurrence, mostly via vibration-related quantities due to damage-induced rattling. However, when the operational environment is highly variable and the online data suffer large uncertainties, robust early detection becomes challenging. Beyond simple detection, however, in-situ prognosis is even of greater interest, since it seeks to determine the remaining useful life (RUL), given condition monitoring data. Under realistic conditions, the nominal operation life, usually in terms of L10, is not practical when the prognosis is subject to the aforementioned uncertainties. This paper aims at updating the most plausible model parameters to obtain an accurate failure curve. From this Bayesian model updating process, the prediction of RUL is therefore made, associated with posterior probability. Vibration data are collected from the SpectraQuest machinery fault simulator, with gradually deteriorating bearings subject to fine foreign particles in the lubrication.