CNN-BiLSTM Hybrid Framework Assesses Aircraft Engine Remaining Service Life

In order to overcome the shortcomings of traditional remaining useful life (RUL) prediction methods, especially their dependence on artificial feature engineering and limited prediction accuracy, this study proposes an end-to-end CNN-BiLSTM deep learning architecture. The method first preprocesses multi-dimensional sensor data, including sliding window segmentation, Z-score normalization, and setting the upper RUL limit to 120, so as to generate high-quality input. The model first uses a two-dimensional convolutional neural network (2D-CNN) to extract local spatial correlations between sensors, and then models bidirectional long-range dependencies in the time series through a bidirectional long-term memory network (BiLSTM). Finally, the fully connected layer outputs RUL prediction results. Experiments on the NASA FD001 dataset show that the model achieves RMSE of 15.65 and MAE of 11.81, which is significantly better than baseline models such as CNN, DCNN, RNN and BiLSTM, with the highest improvement accuracy of 52.09%. Especially at the end of equipment life, it has outstanding performance and shows good engineering application prospects.