Image-data-driven simulation of fluid dynamics (proposal and evaluation)

抄録

Numerical simulation methods driven directly from images have advanced considerably. Oshima (2023, 2024) formulated an immersed boundary Navier–Stokes (IB-NS) equation that treats the solid boundary as a parameter called porosity. This method suggests that flow simulation is driven directly by image-based data without generating surface models. However, specific methods for this approach have not yet been proposed and implemented. Therefore, in this study, we applied a primarily filtering-based image processing technique to calculate the level-set of geometry shapes from image luminance values to apply it to the IB-NS. Using this method, a uniform flow analysis around two-dimensional circular, square and triangular cylinder was performed. The root mean square error of the scalar fields (flow velocity and pressure) was used to compare the calculations based on the porosity generated from the numerical model and the filtered calculations, verifying geometry-specific filter effects. Additionally, the streamlines were compared with a good agreement of the velocity fields. These results confirm that consistency was achieved, and the numerical model can be replaced by filtering. Moreover, actual flow simulations were performed using a two-dimensional RGB image and problem specific to using real images was discussed, along with boundary thickness. Finally, we extended the diffusion filter for calculating the level-set to a three-dimensional binarized voxel-based data and found that it can also be applied to three-dimensional voxel-based data. Therefore, physically consistent numerical solutions were obtained stably, proving that the flow simulation can be performed directly from images without the intermediate step of generating surface models.