Bearing health condition prediction using deep belief network

Abstract

Bearings play a critical role in maintaining safety andreliability of rotating machinery. Bearings health conditionprediction aims to prevent unexpected failures and minimizeoverall maintenance costs since it provides decision makinginformation for condition-based maintenance. This paperproposes a Deep Belief Network (DBN)-based data-drivenhealth condition prediction method for bearings. In thisprediction method, a DBN is used as the predictor, whichincludes stacked RBMs and regression output. Our maincontributions include development of a deep leaning-baseddata-driven prognosis solution that does not rely on explicitmodel equations and prognostic expertise, and providingcomprehensive prediction results on five representative runto-failure bearings. The IEEE PHM 2012 challenge datasetis used to demonstrate the effectiveness of the proposedmethod, and the results are compared with two existingmethods. The results show that the proposed method haspromising performance in terms of short-term healthcondition prediction and remaining useful life prediction forbearings.