基于卷积神经网络的火箭冲压组合发动机燃烧流场重构

A reconstruction model of combustion flow field based on convolutional neural network architecture is proposed. The objective is to reconstruct the two-dimensional temperature field of rocket based combined cycle(RBCC)engine with complex flow field characteristics from a low-resolution temperature field. The turbulent combustion flow field dataset of four different structures of combustors was obtained by a large eddy simulation method. Three of these structures were used as training dataset，and the reconstruction results of the temperature field for the remaining structure was analyzed in order to validate the reconstruction neural network model. The results demonstrate the effectiveness of the temperature field reconstruction model in accurately reconstructing the two-dimensional high-resolution temperature distribution from the low-resolution temperature field. The average error of the temperature field reconstruction in the primary combustion region at the trailing edge of the central rocket is below 5%，surpassing the accuracy achieved by the bicubic interpolation algorithm. The dataset and model presented in this study provide a foundation for the subsequent development of intelligent perception and regulation of combustion for the RBCC engine.