Comprehensive Analyses on the Influential Factors in Supersonic Combustion Simulation Using Dynamic Adaptive Chemistry Method

Abstract

ABSTRACT To overcome the major challenge of reactive flow simulation for chemical kinetics dominated flame dynamics in supersonic combustion, on-the-fly mechanism reduction for high fidelity simulation of scramjet becomes mandatory. For dynamic adaptive chemistry (DAC) methodology, there are three major factors controlling the accuracy and efficiency of the overall simulation, namely, the mechanism reduction method, error threshold value , and search initiating species (SIS). In the present work, systematic investigations of the three influential factors were conducted for large eddy simulation of ethylene-fueled supersonic combustion within a unified DAC framework. The results show that all the four mechanism reduction methods, i.e., DRG, DRGEP, PFA, and DAC-L, are adequate for the combustor’s global performance prediction regarding the wall pressure, stable combustion productions, and temperature. However, for intricate flame stabilization characteristics, the DRG, DRGEP, and DAC-L methods yield comparable prediction accuracy in radical distributions, whereas the PFA method leads to relatively large discrepancies compared to direct integration with the detailed mechanism. The DRGEP method obtains the best balance between numerical accuracy and computational efficiency among the four methods, while the PFA method is the most computationally demanding one. Regarding the mechanism reduction error threshold value, the relative errors in physical property predictions increase as the relaxation of the error threshold value. And the comparative study suggests that the should not exceed for high fidelity simulations of supersonic combustion. Furthermore, the stable species combination, namely, fuel, O2, and N2 incurs larger relative errors in radical mass fraction prediction than the combination including fuel and intermediate species HO2 and CO. Nevertheless, the latter is less computationally efficient than the former as it requires 15% more CPU time to solve the stiff ODE system of the resultant skeletal mechanism. It should be noted that the computational overheads for mechanism reduction under various values and SIS combinations are almost the same, and the overall computational efficiency is mainly determined by the CPU time for solving the stiff ODE system of the size-reduced skeletal mechanisms.