Prediction of remaining useful life of multi-stage aero-engine based on clustering and LSTM fusion

Highlights: • A novel multi-stage ILSTMC model of RUL prediction is proposed. • The clustering algorithm is presented for multi-stage real-time data analysis. • A new prediction algorithm is developed to achieve better prediction performances. Accurately predicting the Remaining Useful Life (RUL) of an aero-engine is of great significance for airlines to make maintenance plans reasonably and reduce maintenance costs effectively. Traditional single-parameter and single-stage models achieve low prediction accuracy. In order to improve the prediction accuracy of the RUL of the aero-engine, a novel aero-engine RUL prediction model named Improved multi-stage Long Short Term Memory network with Clustering (ILSTMC) is proposed. Based on this model, we research a corresponding multi-stage RUL prediction algorithm, which integrates the advantages of clustering analysis and LSTM model. The National Aeronautics and Space Administration (NASA) dataset is adopted for verification. The experimental results show that the method provided in this paper reduces the prediction error of the aero-engine RUL effectively. In the cases of multi-stage prediction, the prediction error of ILSTMC is the smallest compared with LSTM, Recurrent Neural Networks (RNN) and Linear Programming (LP) methods. In the multi-stage prediction of RUL, it is evaluated adopting Root Mean Squared Error (RMSE) and prediction error. The RMSE of the last stage is reduced by 0.85% compared to LSTM, the RMSE of each stage is reduced by 1.87% compared to LSTM on average; the accuracy of life time cycle is better than LSTM by 0.59%, and the average accuracy of life time cycle at each stage is improved by 1.84% compared to LSTM. The results reveal that the proposed ILSTMC model effectively improves the prediction accuracy of RUL.