An Empirical Study of Remaining Useful Life Prediction using Deep Learning Models

Abstract

An accurate remaining useful life (RUL) prediction is critical for successful operation of a target system. In pursuing a better method for RUL prediction, in this paper, we consider three well-known deep learning models: long short-term memory (LSTM), Transformer, and denoising autoencoder (DAE). In particular, we conduct empirical study with various combinations of the three deep learning models. The experiment results first show that DAE is useful to remove noise from the raw input data. The experiment results also show that the combination of DAE and Transformer leads to the best performance.