Turbofan Engine's RUL Prediction based on the CSI-EMD and Double-Channel Multilayer Feature Fusion Network

In this paper, a hybrid network framework based on empirical mode decomposition improved by cubic spline interpolation (CSI-EMD) and double-channel multilayer feature fusion network (DCM-FFN), has been proposed to improve the accuracy of Remaining Useful Life (RUL) prediction. The CSI-EMD is an empirical mode decomposition (EMD) method that we have improved, which decomposes the multi-sensor time series into a bunch of intrinsic mode functions (IMFs), and then the DCM-FFN predicts the concrete states and summarizes the final RUL prediction value. Our proposed CSI-EMD method successfully alleviates the endpoint effect problem in traditional EMD methods. In order to improve the ability of neural network to extract degraded signals, a method combining multi-scale convolutional neural networks (MSCNN) and echo state network (ESN) is adopted in the framework. The proposed approach is evaluated by aircraft turbine engine data from NASA (FD001-FD004). Compared with existing state-of-the-art methods, the root mean square error (RMSE) and Score of the proposed method decreases 15.33% and 54.86% respectively. Therefore, results and comparisons show that the prediction performance of the proposed method has been improved greatly