Frequency Transfer Model: Generating High Frequency Components for Fluid Simulation Details Reconstruction.

In this paper, a novel method is proposed for data-driven high frequency components generation of velocity fields in fluid simulation. It targets on fluid simulation based on N-S Equation which may suffer from details missing because of low simulation grid resolution and other reasons causing energy dissipation. A frequency transfer model structured with deep learning methods is designed to generate high frequency components with low frequency components as inputs. Considering that high frequency components cannot be inferred from lower ones exactly, our model provides another freedom to adjust the ratio of low frequency and high frequency components in velocity fields to obtain results with diversity and reality. A series of evaluations and results are presented to show our model’s effectiveness in generating high frequency components from low frequency components without improving grid simulation to alleviating energy and dissipation and reconstruct details in fluid simulation.