A Swin-Transformer-based model for super-resolution reconstruction of turbulent flows

Obtaining highly accurate turbulent flow fields is a challenging and resource-consuming process in research tasks and practical engineering applications. In this paper, we propose a Swin-Transformer-based model called Multi-Scale Swin Transformer (MSST) to reconstruct high-resolution flow fields by learning features from low-resolution fields. The hierarchical architecture of MSST enables the model to capture features at different levels, making it well suited for multi-scale feature extraction. Forced isotropic turbulence and turbulent channel flow are used as datasets. A loss function based on physical constraints is embedded in MSST, which improves the super-resolution reconstruction accuracy. The reconstructed instantaneous flow fields are comprehensively analyzed and compared. The results show that MSST performs well in the evaluation metrics and can reconstruct the turbulent flow field with high resolution in complex flow field situations, especially on turbulent channel flow.