Optimization of supersonic combustor configuration based on Gaussian process regression and genetic algorithm

Abstract

The present study employs a genetic algorithm to conduct an optimization research on the geometric configuration of the combustor of a supersonic combustor with forward fuel injection at the bottom of the cavity, based on a surrogate model established using numerical simulation and Gaussian process regression. A numerical simulation model for supersonic flow combustion is established, which includes RANS equations, SST turbulence equations, a four-step simplified gas-phase reaction kinetics model for H2/CO/CH4, and a three-step simplified solid-phase surface chemical reaction kinetics model for carbon particles. Through this numerical simulation model, 432 simulations are conducted on a combustor of a solid rocket scramjet considering five design variables to obtain a database. The database is divided into 90 % training set and 10 % validation set. A square exponential kernel function-Gaussian process regression model is used to train the training set and obtain the surrogate model, which is then evaluated using three metrics to assess its accuracy. Finally, genetic algorithm optimization research is conducted on the surrogate model with maximum total pressure recovery coefficient as the objective. The results show that compared to the base case, the optimized case achieves a 15.2 % increase in total pressure recovery coefficient and a 217 % increase compared to the case with lowest total pressure recovery coefficient.