Prediction of performance characteristics of an experimental micro turbojet engine using machine learning approaches

The continuous growth of the world population leads to an increase in energy demand, which poses challenges to sustainable energy supply. Predicting aviation engine performance according to its own characteristics is very important in ensuring sustainability. Moreover, as aviation engines are used in more sectors and for more purposes, it is becoming more crucial to forecast aircraft engine parameters based on their inherent properties. In this study, thrust, exhaust gas temperature (EGT) and specific fuel consumption (SFC) of micro turbojet engine (MTJ-E) generating thrust of 92 N are predicted using Long-Short Term Memory (LSTM) and Support Vector Regression (SVR), where fuel flow, air mass flow, exhaust gas velocity, compressor inlet and outlet pressures and turbine RPM are determined as model inputs. According to experimental results, thrust changes between 9 N and 92 N whereas EGT varies between 503 °C and 613 °C. Moreover, SFC is observed between 0.178 kg/Nh and 0.456 kg/Nh. The findings of performance modeling indicate that the coefficient of determination (R2) for the thrust, EGT and SFC of the MTJ-E is obtained 0.989603, 0.864536 and 0.983209 by SVR, respectively, while the LSTM approach leads these values to enhance 0.999227 for thrust, 0.869209 for EGT and 0.985693 for SFC. On the other hand, mean absolute percent error (MAPE) values for these metrics change from 9.7435 % to 1.7112 % for thrust, from 1.3818 % to 1.3049 % for EGT and from 3.2147 % to 2.4933 % for SFC. For novel engine designs, it could be helpful to model performance metrics by using machine learning with low error, which enables the prediction of interim values.