Prediction method of remaining useful life of aero-engine based on long sequence

A method for forecasting the remaining useful life of an aero-engine based on a stack-dilated convolution neural network (SDCNN) was presented in order to address the inadequate prediction accuracy of the engine's useful life with long-sequence data from many sensors.The multi-sensor long sequence data was normalized to eliminate errors caused by different dimensions and value ranges. A prediction objective function was constructed to represent the real degradation of the aero-engine. A degradation prediction model was built, based on SDCNN, and long-term, deep, and global time series features were extracted by expanding the receptive field of the model for regression analysis, and then the remaining useful life prediction result of aero-engine was obtained.The model's hyperparameters were optimized using the Hyperband optimization algorithm and the StratifiedKFold cross-validation method to increase prediction accuracy and adaptability under various conditions. The commercial modular aero-propulsion system simulation (C-MAPSS) dataset was used to confirm the efficacy of the suggested method. The experimental results based on the FD003 dataset in C-MAPSS show that the proposed method can effectively improve the prediction accuracy of aero-engine remaining life based on long-sequence signals, and the score index to evaluate the prediction accuracy of the model is significantly reduced by 32.62%. © 2024 Beijing University of Aeronautics and Astronautics (BUAA). All rights reserved.