Remaining Useful Life Prediction of Aero-Engine Based on Deep Convolutional LSTM Network

Remaining useful life (RUL) prediction of aero-engine is a crucial problem in the aerospace industry and it is usually based on monitoring sensor parameters. Accurate RUL prediction is of great significance to ensure the safety of aero-engine and save maintenance costs. LSTM is widely used in solving RUL prediction problem because of its obvious advantages on dealing with time series. In order to both consider temporal and spatial features, we use deep convolutional LSTM (ConvLSTM) as basic computation unit. In this paper, we propose a novel method by embedding multilayer ConvLSTMs into U-Net structure. We firstly get the two-dimensional spatiotemporal features by data preprocessing layer and then feed them into the network. In addition, we also provide a simplified model by replacing ConvLSTM with ConvJANET (convolutional Just Another Network), which has less parameters and faster inference speed. Experiments on dataset C-MAPSS are conducted and the proposed model shows the advantages both in effectiveness and speed compared with other related methods.