A hierarchical scheme for remaining useful life prediction with long short-term memory networks

Abstract

Remaining useful life (RUL) prediction is essential in prognostics and health management (PHM) applications, where data-driven approaches employ the tendency of the degradation process using operating data of complex systems, and have attracted more and more attention. With the idea that forecasting the time period before the equipment reaches the critical degradation stage (e.g., failure, fault, etc.), RUL prediction is usually formed as an optimization problem (in particular, a regression problem between the inputs–real-time measurements and the outputs–the RUL predictions). This work formulates the RUL prediction as a bi-level optimization problem, (i) the lower level is intended to forecast the time-series in the near future, and (ii) the upper level is to predict the RULs by integrating the available measurements up-to-date and the predicted ones by the lower-level prediction. To tackle the hierarchical optimization problem, a bi-level deep learning scheme is proposed for the machine RUL prediction, where long short-term memory (LSTM) networks are applied as of the unique characteristics in processing time-series and extracting recursive and non-recursive features among them. Case studies using PHM08 data challenge data set, 4 data sets in C-MAPSS package and 1 data set in the new CMAPSS dataset are implemented, to validate the proposed framework. The results show that the presented method outperforms the state-of-the-art approaches.