Prediction of Bearing Service Life Using an Auto Regression Moving Average and Response Surface Methodology

Abstract

Accurate prediction of service life of machines and their components is very important for reliability evaluation, and efficiency. It is intended in the current work to introduce a method to predict service life of bearings using accelerated life test rig. This is based on the measurement of vibration signals during an accelerated life test and applying an Auto Regressive Moving Average (ARMA) technique together with the Response Surface Methodology (RSM). Vibration signals are measured using accelerometers attached to deep groove ball bearings supporting a rotating shaft. Recorded signals are fed offline to a time-series routine within the SPSS software package where the coefficients of the ARMA models are estimated. Such models are usually utilized to predict the level of vibration signals, which is directly related to service life. The developed models are adequate enough to provide reasonable predictability measures. It is generally found that for all sets of data, the current response value is influenced by the past history of the preceding impacts and the residuals moving average. For better and comprehensive indication of the real functional interrelations between service lifetime of the tested bearing and the measured vibration signals, experimental signals amplitudes are represented in 3D surfaces and contour maps. It is found that signals amplitudes are drastically magnified near the end of bearing service life.