Remaining useful life prediction of rolling bearing via composite multiscale permutation entropy and Elman neural network

Abstract

Accurate prediction of remaining useful life (RUL) is very important for the maintenance of mechanical equipment. With the latest development in sensor technology and artificial intelligence, the data-driven prediction method of mechanical equipment RUL has been widely concerned. However, in the past, the research on data mining was not sufficient, and the method of extracting degradation features was single. In order to solve this problem, this paper proposes a new method to extract the degradation features of bearings. This method creatively combines image with entropy methods and uses the Symmetrized Dot Pattern (SDP) method and Composite Multiscale Permutation Entropy (CMPE) to extract degradation features. Through five evaluation indexes, the degradation feature extracted by the SDP-CMPE method is compared with five common degradation features in time domain. Finally, a simple Elman neural network is used to predict RUL. Through the case study of bearing data set, it is verified that the degradation features extracted by the SDP-CMPE method are superior in prediction accuracy and conservatism.