Bearing Remaining Useful Life Prediction Based on NAP-HSMM

Abstract

The prediction of remaining useful life (RUL) is the most challenge task in the area of reliability and safety of rolling bearing. Nuisance attribute projection (NAP) is used in this paper to eliminate the impact of operating conditions in vibration signals, as the inevitable errors in the manufacturing, installation, or operation of the bearings. Due to the limitation of hidden Markov model (HMM), an HSMM with aging factor is designed to solve problem of RUL prediction. Time-domain and other features are extracted from vibration signals and then transformed by NAP projection so can be used to train HSMM model. When the health state is determined by HSMM, state remaining duration estimation is used to help calculation of RUL. In order to test the prediction ability of the model, PHM 2012 challenge data are used, the result showed that the HSMM with NAP has made a great improvement than ones without.