Abstract:

Bearings are critical components employed virtually in all rotating machines and automobiles to alleviate friction between surfaces during relative motion. In traditional approaches, rolling element bearing failures are predicted based on either historical time-to-failure data (event data) or condition monitoring (CM) data. Prediction methods using event data are of little value to maintenance decision making since they render general forecasts for the total population of identical units instead of forecast for a particular unit presently operating in the machine. Prognosis based on CM data provides short term predictions which may not be useful in maintenance scheduling. Proportional hazards model (PHM) can be used to predict hazard rates and reliability of machines and its components using both event data and CM data.
This paper presents a method for defect prognosis of roller bearings using Weibull proportional hazards model (WPHM) based on parameters obtained from vibration analysis and historical event data. Morlet wavelet filter (MWF) is used for denoising of vibration signals. Time domain parameters extracted from the denoised vibration signals are used as covariates in the WPHM. Use of log-likelihood parameters as covariates in WPHM is explored and their performance is compared with that of other parameters. The proposed approach helps in early estimation of hazard and reliability with more accuracy, eventually increasing the effectiveness of condition based maintenance and reducing maintenance costs.
Received on August 31, 2009, revised on March 17, 2010
References: 20