Aeroengine Gas Path Parameter Trend Prediction Based on LSTM

Accurately predicting the trend of aero-engine gas path parameters is crucial for ensuring safe flight and enabling condition-based maintenance. However, the demanding and uncertain service environment introduces challenges in dealing with the noisy and non-stationary data collected by engine gas path sensors. Traditional time series models struggle to accurately predicts parameter trends, resulting in insufficient fitting and prediction accuracy. In this paper, we address these challenges by leveraging the characteristics of engine post-flight data and introducing Long Short-Term Memory (LSTM), a type of artificial neural network in deep learning. We construct both single-feature input and multi-feature input LSTM prediction models for six key indicators of engine gas path performance. We analyze the models' capabilities for single-step and multistep predictions. To evaluate the effectiveness of our approach, we compare the LSTM model with the traditional Autoregressive Moving Average (ARMA) model and support vector regression (SVR) method. The results demonstrate that the LSTM model outperforms the traditional ARMA and SVR models in terms of prediction accuracy and stability. This indicates that utilizing LSTM is an effective approach for improving the accuracy of engine gas path parameter prediction. By accurately predicting these parameters, we can enhance flight safety and enable more efficient condition based maintenance.