Development of 3D Molecular Tagging Velocimetry with Fourier Integral Microscopy and Neural Radiance Fields
B. Thurow(1), S. Halder(1), B. Sapkota(1), M. Yamakaitis(2), C. Fort(2), P.M. Bardet(2)
1: Department of Aerospace Engineering, Auburn University, USA
2: Department of Mechanical and Aerospace Engineering, The George Washington University, USA
DOI:
Tomographic reconstruction of three-dimensional flow fields from limited angular views is a fundamental challenge in experimental fluid diagnostics. Conventional grid-based iterative methods amplify noise and produce reconstruction artifacts under the narrow angular sampling conditions imposed by compact imaging systems, and remain computationally expensive. Neural Radiance Fields provide a continuous, gridless reconstruction framework; however, conventional NeRF implementations based on Fourier encoding typically require large network architectures, resulting in a computationally expensive reconstruction process. In this work, the FluidNeRF framework was adopted and modified to support limited-angle orthographic ray formulation for MTV applications. A method was developed for synthetic data generation for the purpose of performance characterization. The results establish the modified FluidNeRF framework as a computationally efficient and viable alternative to classical tomographic approaches for time-resolved MTV reconstruction.
