Remaining Useful Life Estimation in Prognostics Using Deep Bidirectional LSTM Neural Network

Abstract

Remaining Useful Life (RUL) estimation plays a crucial role in Prognostics and Health Management (PHM). Traditional RUL estimation models are built on sufficient prior knowledge of critical components degradation process which is not easily available in most situation. With the development of integrated circuit and sensor technique, data-driven approaches show good potential on RUL estimation. This paper proposes a new data-driven approach with Bidirectional Long Short-Term Memory (BiLSTM) network for RUL estimation, which can make full use of the sensor date sequence in bidirection. By visualized analysis of the hidden layers, the model can expose hidden patterns with sensor data of multiple working conditions, fault patterns and degradation model. With experiment using C-MAPSS dataset, BiLSTM approach for RUL estimation outperforms other traditional approaches for RUL estimation.