An intelligent prediction method for supersonic flow field in scramjet isolator enhanced by feature details

Abstract

The accurate and fast prediction of hypersonic flow field can provide reliable data for the analysis of flow evolution process. The traditional unsteady numerical simulation method requires a large amount of time and economic cost to carry out flow prediction. The flow field prediction algorithm based on deep learning has been proved to be an effective modeling tool and approximator. In order to predict the complex flow process of scramjet isolation segment with high precision, an intelligent prediction model combining position coding and detail feature recovery is proposed. Positional encoding is used to capture the spatial distribution characteristics of pressure points, while multi-scale convolutions within the encoder-decoder framework extract multi-scale features from different receptive fields. This enhances the model's capability to capture detailed features of the Mach field. The method is validated with data from unsteady numerical simulations using various suction parameters. Experimental results show that this method effectively recovers characteristic parameters such as the position and area of the separation zone in the Mach field. Compared to the Neural Network model of Multipath Fusion (MBFCNN), the proposed method improves the Structural Similarity Index (SSIM) by 6.71% and the Peak Signal-to-Noise Ratio (PSNR) by 44.33% on the test dataset.