Life prediction of jet engines based on LSTM-recurrent neural networks

Abstract

The issue of remaining useful life (RUL) prediction has already become a quite interesting topic in industrial product. The data driven RUL prediction has been applied to the current research by taking advantage of a long-short term memory (LSTM)-recurrent neural network (RNN) approach. This means that even in a specified long-short term memory bound and limited available data sets, the RUL predictions can also improve the equipment capacity. By collecting the sensor parameters from National Aeronautics and Space Administration (NASA) jet engines, the capability of this approach has been demonstrated. An appropriate selection, that is suitable for the variable measurement, has been used to feed LSTM-RNNs with the most useful RUL labels. The analysis results illustrate that, compared with traditional statistical model, the development prediction approach is likely to provide an accurate prediction for RUL equipment that can be meaningful to maintenance schemes.