Remaining Useful Life Prediction Based on a Bi-directional LSTM Neural Network

Abstract

Electric motors have been widely used in the fields of national economic construction, scientific research, medical treatment and national defense. The health of motors plays key role in ensuring the safety of these fields, however, the online health monitoring of motors is not well studied. On the other hand, the combination of health science and artificial intelligence technology is playing an increasingly important role in replacing the traditional health monitoring of machines and has been proved its ability in serial data processing and other aspects. In this paper, a bi-directional cyclic neural network based algorithm is proposed for the intelligent remaining useful life (RUL) prediction of motors. Compared with the traditional one-way neural network, bi-directional cyclic neural network can predict the current state based on the past and future information at the same time, which obtains higher accuracy. This paper is organized in two stages: first, a health index is developed to fit the life cycle data of motors; Secondly, a bi-directional cyclic neural network based model is trained based on the health index for the online RUL prediction of motors. The simulation results show the effectiveness of the proposed method.