Aeroengine remaining useful life prediction using an integrated deep feature fusion model

Aeroengine plays a significant role in advanced aircrafts. Predictive maintenance can enhance the safety and security, as well as save amounts of costs. Remaining useful life (RUL) prediction can help make a scientific maintenance schedule. Therefore, an integrated deep feature fusion model is proposed for aeroengine RUL prediction. Firstly, a nonnegative sparse autoencoder (NSAE) is applied for unsupervised deep feature fusion. Secondly, gated recurrent unit (GRU) is stacked upon the NSAE for temporal feature fusion to model the aeroengine degradation process by its powerful long term dependency learning ability. Finally, an integrated deep feature fusion model with NSAE and GRU is globally finetuned for RUL prediction. A simulated turbofan engine dataset is used to verify the effectiveness, and the results suggest that the proposed method is able to accurately predict the RUL of each test unit.