A Comparative Study of CNN-Based Predictive Maintenance Techniques on C-MAPSS Dataset

In this study, convolutional neural network (CNNs) based predictive maintenance ($\mathbf{P d M}$) techniques are performed on the well-known jet engine data called the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS). We first employ the one-dimensional CNN (1D-CNN) for predicting the remaining useful life (RUL) of the jet engines. Then, to further improve the prediction performance, we combine the long shortterm memory (LSTM), bidirectional LSTM (BiLSTM), and gated recurrent unit (GRU) layers with the 1D-CNN. The performances of all CNN architectures are comparatively evaluated in terms of root mean square error (RMSE), training time, and correlation coefficient metrics. The simulation results show that the integration of LSTM, BiLSTM, and GRU layers with 1D-CNN improves the RUL prediction performance in terms of both correlation coefficient and total RMSE, compared to the classical 1D-CNN model.