This study proposes a fluid analysis system that automatically acquires geometric data from images, eliminating manual model generation for objects with complex geometries. Numerical simulation of natural fluid phenomena is essential in many applications; however, the lack of pre-existing geometric models, such as 3D CAD models, remains a major obstacle in natural environments. To overcome this limitation, we leverage recent advances in photogrammetry, particularly machine-learning-based photogrammetry that reconstructs smooth 3D geometry directly from images. Despite growing interest in integrating photogrammetry with numerical analysis, applications of photogrammetry to fluid simulation remain limited. We develop a fluid analysis framework that directly incorporates neural implicit representations into an immersed boundary formulation based on Nitsche’s method. To accelerate the computation, we employ distributed-memory parallel computing. The proposed system is validated through simulations of flow around a circular cylinder, demonstrating accurate reproduction of the flow field using high-resolution Cartesian grids. A demonstration example further illustrates the end-to-end workflow from image-based reconstruction to fluid simulation.
NEMOTO et al. (2026) studied this question.