A Comparative Analysis of Exhaust Gas Temperature Based on Machine Learning Models for Aviation Applications

Abstract

The main objective of this study is to investigate elaborately the relationship between exhaust gas temperature (EGT) and various operational parameters specific to aero-engine for the cruise phase. EGT prediction is performed based on different models, including deep learning (DL) and support vector machine (SVM), using a set of historical flight data, more than 1300. In order to achieve this goal, the EGT is taken as the output parameter while the most key variables for the EGT prediction are taken as the input parameters to the models. Several statistical goodness tests, namely root-mean-square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2), are conducted to make a fair comparison between the efficiency and performance of each model that is developed based on Matrix Laboratory (matlab) and R code. The relative importance for the altitude (ALT) parameter of 11.89% has the highest value while the lowest relatively importance parameter is vibration (VIB) of 5.00%. EGT variation for the actual data is in the range of 459.05 and 607.32 °C. It is observed that the EGT variation of DL and SVM ranges from 457.09 to 604.52 and from 454.64 to 603.23 °C, respectively. Furthermore, the prediction error for DL and SVM fluctuates between a minimum of −21.61 to a maximum of 22.50 °C and a minimum of −13.34 to a maximum of 12.44 °C, respectively. In the light of the statistical test results, it is concluded that the DL model with RMSE of 4.3922, MAE of 3.3981, and R2 of 0.9834 shows more excellent ability in predicting EGT than the SVM model with RMSE of 5.5212, MAE of 4.0527, and R2 of 0.9712. This study may effectively be applied to different aircraft types as a useful roadmap for academic and industrial researchers in this sort of application and it shed the light on optimizing performance for a specific aircraft by thermodynamic methods.