A New Approach for Remaining Useful Life Estimation Using Deep Learning

Prognosis and Health Management (PHM) refer specifically to the prediction phase of the future behavior of the system or subsystem, including the remaining useful life (RUL). It is helpful to early detect incipient failures in many domains as aircraft, nuclear reactor, turbine gas, etc. In this paper we propose a new approach based on the implementation of data-driven methods for fault prognosis. Such methods require the availability of data describing the degradation process; when there is a lack of data, it is difficult to predict the states using deep models, which require a large amount of training data. In this paper, we propose to use a simple data augmentation strategy to solve the problem of data scarcity in prediction of RUL provided through the use of a long-short term memory (LSTM), which is a type of recurrent neural network. The results of our experiments demonstrate that using a simple data augmentation strategy can increase RUL prediction performance by using LSTM technics. We analyze our approach using data from NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS).