Parallel Attention Network using Vector with High Correlation with Label for Remaining Useful Life Estimation

Ye-In Park, Jouwon Song, Suk-Ju Kang

Keywords: Prognostic health management, remaining useful life, time series analysis, attention network, deep learning, hybrid model

TL;DR: Attention network for remaining useful life prediction on time series

Abstract: Prognostic health management (PHM) has become important in many industries as a critical technology to increase machine stability and operational efficiency. Recently, various methods using deep learning to estimate the remaining useful life (RUL) as a core task of PHM have been proposed. However, the existing methods do not explicitly capture the correlation between temporal and spatial time series, reducing the RUL prediction accuracy. This paper proposes a novel RUL prediction algorithm using a spatio-temporal attention mechanism to based on the vector highly correlated with label to solve this problem. The proposed model constructs three paths in parallel, a time-oriented attention network, a feature-oriented attention network, and a bidirectional long short-term memory (LSTM) network. The first two attention networks focus on temporal and spatial information required for RUL prediction based on convolutional neural network (CNN), respectively. Unlike existing attention networks, the proposed attention network uses the vector learned in the intermediate prediction process as a query vector to focus on time series data related to the RUL. The last bidirectional LSTM network is additionally configured to compensate for the inability of the CNN-based attention networks to grasp continuous time distributions. Experiments have been performed on two widely used datasets and experimental results show that the proposed approach outperforms the state-of-the-arts.