Physics-Aware Spatio-Temporal Dynamics and Test-Time Refinement for Turbulent Flow Reconstruction.

A new physics-guided neural network is developed for reconstructing the full-scale DNS field from low-resolution LES data in turbulent flows. The method utilizes the transport equations that underlie the flow dynamics to design the spatio-temporal model architecture. A degradation-based refinement method is also developed to enforce physical constraints, and to reduce accumulated reconstruction errors over long periods. The model is shown to reconstruct the long-term continuous spatial and temporal dynamics of the flows in an accurate manner. Data from two incompressible turbulent flow configurations are used to evaluate the performance of the model and to compare it with previous super-resolution models. Detailed qualitative and quantitative comparative assessments demonstrate the effectiveness of each of the components of the new model.