Prediction of aircraft engine parameters from real sensor values using machine learning throughout flight envelope

In recent years, the prediction of aircraft engine parameters using machine learning methods has become increasingly common and widely preferred. The forecasting of aircraft engine parameters based on actual flight data is of great importance in terms of maintenance facilitation and safety. In this study, fuel flow, exhaust gas temperature, and N1 shaft RPM are predicted based on real-time recorded aircraft and engine data over a flight envelope. In this regard, three different machine learning algorithms (ridge regression, extreme gradient boosting (XGBoost), and neural basis expansion analysis for time series (N-BEATS)) are employed to predict these parameters. As input parameters, flight speed, angle of attack, thrust resolver angle, total air temperature, compressor outlet pressure, N2 shaft speed, and pressure altitude were considered. When considering the results, it is observed that aircraft flight speed has the highest importance in predicting exhaust gas temperature and N1 shaft speed, with importance scores of 19.74% and 16.56%, respectively. On the other hand, for fuel flow prediction, compressor outlet pressure has the highest importance at 20.71%. Furthermore, using the ridge regression method, EGT, fuel flow, and N1 shaft speed are predicted with R² values of 0.9223, 0.9893, and 0.9961, respectively, whereas these are forecasted with R² values of 0.9349, 0.9759, and 0.9899 by N-BEATS. Moreover, these engine parameters are further improved by using the XGBoost method, reaching R² values of 0.9980, 0.9974, and 0.9996, respectively. Finally, it is concluded that engine parameters could be accurately predicted from aircraft parameters with relatively low error values.