A deep attention residual neural network-based remaining useful life prediction of machinery

Abstract

Remaining useful life (RUL) estimation has always been an essential task of prognostics health management (PHM). However, degradation data of machinery is seldom available, which limits the performance of data-driven models. Besides, the prediction performance of the established model may also be awfully affected by working environment of monitored machinery. To over these problem, a novel deep attention residual neural network (DARNN) is proposed by us for RUL prediction of machinery. The proposed DARNN has following advantages: (1) Representations of degradation can be effectively extracted from signals by the proposed DARNN. (2) The prediction performance and self-stability of the proposed model significantly surpassed some existing methods. Two case studies are conducted for the RUL prediction of bearings and turbofan engines respectively to verify the effectiveness of the proposed model.