A dual attention LSTM lightweight model based on exponential smoothing for remaining useful life prediction

Abstract

Accurate remaining useful life (RUL) prediction of degrading systems is crucial to predict failures in advance and develop maintenance plans. As systems degrade gradually over time, sequential degradation feature (SDF) is very important. However, in attention mechanism (AM) based RUL prediction approaches, the sequential operation at each time step is abandoned. Further, these methods are modeled based on numerous parameters, making it difficult to enable timely RUL prediction. Therefore, this paper proposes a dual attention and long short-term memory (LSTM) lightweight model (DA-LSTM). LSTM compensates for the shortcomings of AM in modeling SDF, and exponential smoothing is adopted to train a lightweight model. Specifically, the SDF is divided into aggregated encoding feature (AEF) and aggregated original feature (AOF). AEF is obtained by the encoder which includes a novel soft attention mechanism and an LSTM network. To prevent losing useful information during the encoding process, the second attention layer aggregates the original sensor signal to obtain AOF. Finally, the decoder LSTM network combines AEF with AOF and calculates RUL based on a weighting average method. Extensive experiments are conducted on the C-MAPSS dataset to verify model effectiveness. The results show the superiority of DA-LSTM in prediction accuracy and computational quantity.