Non-linear Iterative Optimization Method for Locating Particles Using HPC Techniques

Abstract

Tomography has been recently introduced in fluid velocimetry to provide three dimensional information of the location of particles. In particular, author's previous works have proven the potential of Optical Diffraction Tomography for biological and microfluidic devices. In general, image reconstruction methods at visible wavelengths have to account for diffraction. First Born Approximation has been used for three dimensional image reconstruction, but a non-linear reconstruction method is required when multiple scattering is not negligible. Therefore, to improve the spatial resolution of the fluid velocimetry techniques, a non-linear iterative optimization should be used to locate the seeding particles and compute afterward the flow velocity field. This inversion method requires the solution of the Helmholtz equation, computationally highly demanding due to the size of the problem. Therefore, High Performance Computing is required to find the particle locations. This work shows the results of accelerating this task using GPU computing and a customized storing format.