Supersonic combustion flow field reconstruction in a scramjet based on deep learning method

Abstract

Efficiently predicting combustion flow in scramjet engines enhances early state awareness, which is crucial for facilitating active flow control and ensuring reliable and stable engine operation. Aiming at the problem that the traditional optical test method of ground wind tunnel test is limited by complex and narrow space and difficult to obtain data, the study introduces a supersonic flow field reconstruction model that utilizes local-global feature grouping and fusion to realize fast reconstruction of flow field schlieren image based on sparse pressure data obtained by tiny wall pressure sensor. To prevent the loss of many shallow gradients in the transmission process, a two-layer gradient calculation strategy is designed to preserve the shallow gradient features to the maximum extent. To reduce the number of model parameters and enhance the information crafty interaction between different flow fields, a binary segmentation mask strategy is designed to transform the image dimensionality and block. To assess the effectiveness of the proposed model, experiments were conducted using hydrogen fuel combustion test data obtained from a pulse combustion wind tunnel with an inflow Mach number of 2.5. When compared to other models, our model demonstrated a significant 14.37 % improvement in structural similarity and an 8.35 % improvement in peak signal-to-noise ratio indicators. Most notably, these improvements were achieved while maintaining the lowest computational complexity.