Sparse feature extraction-based flowfield reconstruction and prediction in a supersonic isolator

The reconstruction and prediction of the flowfield within the scramjet isolator serve as the foundational elements for the surveillance and regulation of the power system in the realm of supersonic flight. An important method for doing so accurately is feature extraction. In this study, a sparse feature extraction-based supersonic flowfield reconstruction and prediction methodology is proposed. Ground experiments are performed on a supersonic isolator with variable Mach number and backpressure: sparse features of the flowfield therein are obtained by subjecting experimental schlieren to discrete orthogonal transformation and sparse processing, then reconstruction is accomplished via inverse transformation. To predict the flowfield via only a small number of wall pressure sensors, a sparse prediction model is established by combining a convolutional neural network and a fully connected network: the trained model predicts the sparse features, from which the predicted flowfield is derived. Analysis results show that the flowfield reconstructed from the sparse features has almost all the characteristics of the original one. The reconstruction accuracy exceeds 90% with fewer than 10% of the sparse features, and the predicted sparse features accurately represent the key local ones and the global structure.