Chemical Reactor Network and Agent Model for Hypersonic Strut–Cavity Combustor Flames

The study proposes a generic chemical reactor network (CRN) for a supersonic combustor. Despite the typical criterion of splitting chemical reactors by flow parameters (for example, mixing fraction, heat release rate, velocity, etc.), pressure is added to reflect the pressure fluctuations in supersonic combustion. The temperature prediction of the CRN reached less than 2% error as compared with three-dimensional numerical simulation. An agent model of the CRN was developed to predict local thermodynamic states at high-Mach-number conditions. For sensitive variables introduced by the CRN (for example, heat flow coefficients and volume correction coefficients), Latin hypercubic sampling was applied to obtain the dataset, and a multilayer perceptron model was used for optimization. A backpropagation method was applied to predict the parameters of the CRN for a given outlet condition provided by experiments. Reconstruction of the chemical reaction processes was achieved at low computational cost. Therefore, the methodology developed in this study can accelerate the optimization process of supersonic combustors for hypersonic airbreathing engines.