Remaining Useful Life Prediction for Aero-Engine Based on LSTM and CNN

Data-driven Remaining Useful Life (RUL) prediction for aero-engine has evolved rapidly in recent years. Especially, deep learning-based methods like Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) have achieved excellent results. However, there is still limited study to identify the effect on network performance from the number of convolutional layers, LSTM layers and their combination structure. Therefore, the optimal number of convolutional layers and LSTM layers was first determined for CNN and LSTM respectively in this paper. A combined network CNN-LSTM was then constructed. Three kinds of deep networks (CNN, LSTM and CNN-LSTM) were compared on aero-engine RUL prediction. Experimental results on the C-MAPSS dataset indicated that LSTM with 2 dense layers achieved the highest prediction accuracy.