Modeling of energy and emission metrics of a turboshaft engine with hybrid algorithm involving machine learning and metaheuristic approaches

Abstract

Aviation engines have attracted huge interest due to their applicability across various fields, essentially owing to their high power-to-mass ratio, which is considered a critical performance index. In this study, effects of power setting changing from 0.6 to 1 and water-to-air ratio (WAR) ranging from 0 to 0.05 on core efficiency, shaft power, specific fuel consumption (SFC) and NOx emission regarding turboshaft engine used helicopters are numerically investigated. Based on the data obtained from this analysis, these metrics of the engine are modeled by regression analysis. The models are subjected to optimization methods such as simulated annealing, bat and genetic algorithms. Moreover, the engine parameters are modeled by different machine learning approaches involving K-nearest neighbors, random forest and extreme gradient boosting. Lastly, the hybrid algorithm called Classification and regression tree-Harris Hawks optimization (CART-HHO) is proposed to model these parameters. According to performance evaluations, specific fuel consumption of the engine changes between 0.2749 kg kW-1 h-1 and 0.7397 kg kW-1 h-1 whereas core efficiency of the engine resides between 19.34% and 36.9% due to influences of both power setting and WAR variables. On the other hand, the highest R2 is obtained by the proposed method called CART-HHO method. Namely, R2 of core efficiency is measured as 0.9572 whereas R2 of SFC is obtained as 0.9424 thanks to the hybrid CART-HHO approach. These outcomes show that characteristics of the engine depending on external factors are predicted high accuracy.