Machine learning application to forecasting performance and thermodynamics parameters of small turbojet engine

Abstract

The common use of micro-turbojet engines to meet various needs of the modern world makes these more popular day by day; hence, some scientists have devoted their studies to these engines which enable efficiency improvement of key components. In the design phase, solving relationship between characteristics of gas turbine engines helps determine how the engine demonstrates behavior under the analyzed conditions. To predict performance parameters depending on the variables of engine itself, machine learning approach is one of the best ways to present effective solutions thanks to comprehensive algorithm options. In this study, several parameters such as thrust, exergy efficiency, thermal efficiency and sustainability index of a conceptual micro-turbojet engine are forecasted by employing random forest (RF) and gradient boosting machine (GBM) algorithms. For this aim, four input variables are determined, which are air mass flow, fuel mass flow, compressor pressure ratio and engine pressure ratio. According to performance results, net thrust of the engine changes between 1.58 kN and 3.04 kN, whereas exergy efficiency varies between 5.81% and 14.04% with respect to operation points. As for prediction outcomes, both algorithms could forecast the metrics of engine with high accuracy; however, there is slight difference in prediction success of parameters. Namely, coefficient of determination (R2) of thrust is found as 0.9993 by RF, while it is computed as 0.9987 by GBM. In other words, MAPEs of the metrics are found between 0.0043 and 0.0065 by RF, whereas these are measured between 0.0092 and 0.017 by GBM. These findings show that performance indicators of jet engine could be sensitively estimated from itself-specifications by using RF and GBM approaches.