Multi-Resolution LSTM-Based Prediction Model for Remaining Useful Life of Aero-Engine

Aircraft is an important means of travel and the most convenient and fast vehicle in long-distance transportation. The aircraft engine is one of the most critical parts of an aircraft, and its reliability and safety are extremely important. In this article, we consider that the operating conditions of aero-engines are complex and changeable, and a multi-resolution long short-term memory (MR-LSTM) model is proposed. The model can effectively predict the remaining useful life (RUL) of an aero-engine, which is a priority issue within the Prognostics and Health Management (PHM) framework – and thus it can support maintenance decisions. Sequences with multiple temporal resolutions are generated by a reconstruction of the decomposed wavelets. A two-layer LSTM model is then designed: 1) the first layer LSTM is used to learn attention at different time resolutions as well as to generate an integrated historical representation; 2) the second layer LSTM is used to learn the long and short-term time dependencies in the integrated historical representation. Experimental evaluations using the C-MAPSS datasets (FD002 and FD004) and the N-CMAPSS dataset showed that compared to other state-of-the-art RUL prediction methods, the FD002 sub-dataset showed a 12.1% reduction in RMSE and a 3.8% reduction in Score; the FD004 sub-dataset showed a 21.8% reduction in RMSE and a decreased by 62.1%; the RMSE of the N-CMAPSS dataset decreased by at most 25.8%.