Intelligent prediction for remaining life of aero-engine --A Study on CNN-BiLSTM Model Based on Scores Loss Function

Abstract

Listen

Translate article

With the advent of the era of big data and artificial intelligence, predicting and managing the remaining service life of aero-engines has become a key challenge, which has a direct impact on the enhancement of autonomous control capabilities. In this paper, a new deep learning model combining a one-dimensional convolutional neural network (1D-CNN) and a bidirectional long and short-term memory network (BiLSTM) is proposed to efficiently handle dynamic and uncertain engine health state data. The concurrent model takes full advantage of 1D-CNN in extracting local features and BiLSTM in capturing time-dependence, avoiding the loss in the information transfer process in conventional models. Empirical studies on NASA's C-MAPSS dataset show that the model significantly improves the accuracy and robustness in predicting the remaining service life, reducing the RMSE by up to 5.05% and the Scores by up to 54.89% compared to the conventional model. Especially in the short-term prediction task, the model shows higher stability and accuracy, which provides strong support for advance warning in the field of aviation safety.