Remaining Useful Life Prediction Based on Improved LSTM Hybrid Attention Neural Network

In recent years, data-driven fault prediction and health management (PHM) methods based on sensor data have achieved rapid development. Predicting the remaining useful life (RUL) of mechanical equipment is not only efficient in averting abrupt breakdowns, but also in optimizing the equipment's operating capacity and lowering maintenance expenses. This study proposed a prediction model based on an improved LSTM hybrid attentional neural network to better forecast the RUL of mechanical equipment under multi-sensor conditions. The temporal pattern attention (TPA) module uses the features extracted by the LSTM module to weight their relevant variables and increase the model's capacity to generalize to complex data sets. In comparison to the current mainstream RUL prediction methods, the improved LSTM hybrid attentional neural network has better prediction performance and generalization capability on the turbofan engine simulation dataset (C-MAPSS) after experimental tests.