Remaining Useful Life Prediction via Attention Mechanism-based LSTM Neural Networks

Abstract

Remaining useful life (RUL) prediction is one of the most challenging problems for modern engineering systems, which is of great significance to avoiding catastrophic failures and reducing economic losses. In an effort to make use of multi-sensor monitoring data and enhance prediction accuracy, a novel RUL prediction approach is proposed based on long and short term memory (LSTM) network with an attention mechanism in this paper. By using the input attention mechanism, the proposed network can selectively focus on certain important inputs without any prior knowledge. Experimental results have illustrated that the proposed approach can achieve superior RUL prediction performance compared with other conventional networks.